Skip to content

D
DeepSpeech

PaddlePaddle / DeepSpeech
大约 2 年 前同步成功

通知 210
Star 8425
Fork 1598
D
DeepSpeech
提交 9699c007 编写于 作者: 小湉湉's avatar 小湉湉
浏览文件
操作

change the docstring style from numpydoc to google, test=tts

上级 683679be
展开全部 隐藏空白更改
内联 并排
Showing with 2350 addition and 4150 deletion
paddlespeech/t2s/datasets/data_table.py
浏览文件 @ 9699c007
......@@ -22,26 +22,17 @@ from paddle.io import Dataset
class DataTable(Dataset):
"""Dataset to load and convert data for general purpose.
Parameters
----------
data : List[Dict[str, Any]]
Metadata, a list of meta datum, each of which is composed of
several fields
fields : List[str], optional
Fields to use, if not specified, all the fields in the data are
used, by default None
converters : Dict[str, Callable], optional
Converters used to process each field, by default None
use_cache : bool, optional
Whether to use cache, by default False
Raises
------
ValueError
If there is some field that does not exist in data.
ValueError
If there is some field in converters that does not exist in fields.
Args:
data (List[Dict[str, Any]]): Metadata, a list of meta datum, each of which is composed of several fields
fields (List[str], optional): Fields to use, if not specified, all the fields in the data are used, by default None
converters (Dict[str, Callable], optional): Converters used to process each field, by default None
use_cache (bool, optional): Whether to use cache, by default False
Raises:
ValueError:
If there is some field that does not exist in data.
ValueError:
If there is some field in converters that does not exist in fields.
"""
def __init__(self,
......@@ -95,15 +86,11 @@ class DataTable(Dataset):
"""Convert a meta datum to an example by applying the corresponding
converters to each fields requested.
Parameters
----------
meta_datum : Dict[str, Any]
Meta datum
Args:
meta_datum (Dict[str, Any]): Meta datum
Returns
-------
Dict[str, Any]
Converted example
Returns:
Dict[str, Any]: Converted example
"""
example = {}
for field in self.fields:
......@@ -118,16 +105,11 @@ class DataTable(Dataset):
def __getitem__(self, idx: int) -> Dict[str, Any]:
"""Get an example given an index.
Args:
idx (int): Index of the example to get
Parameters
----------
idx : int
Index of the example to get
Returns
-------
Dict[str, Any]
A converted example
Returns:
Dict[str, Any]: A converted example
"""
if self.use_cache and self.caches[idx] is not None:
return self.caches[idx]
......
paddlespeech/t2s/datasets/preprocess_utils.py
浏览文件 @ 9699c007
......@@ -18,14 +18,10 @@ import re
def get_phn_dur(file_name):
'''
read MFA duration.txt
Parameters
----------
file_name : str or Path
path of gen_duration_from_textgrid.py's result
Returns
----------
Dict
sentence: {'utt': ([char], [int])}
Args:
file_name (str or Path): path of gen_duration_from_textgrid.py's result
Returns:
Dict: sentence: {'utt': ([char], [int])}
'''
f = open(file_name, 'r')
sentence = {}
......@@ -48,10 +44,8 @@ def get_phn_dur(file_name):
def merge_silence(sentence):
'''
merge silences
Parameters
----------
sentence : Dict
sentence: {'utt': (([char], [int]), str)}
Args:
sentence (Dict): sentence: {'utt': (([char], [int]), str)}
'''
for utt in sentence:
cur_phn, cur_dur, speaker = sentence[utt]
......@@ -81,12 +75,9 @@ def merge_silence(sentence):
def get_input_token(sentence, output_path, dataset="baker"):
'''
get phone set from training data and save it
Parameters
----------
sentence : Dict
sentence: {'utt': ([char], [int])}
output_path : str or path
path to save phone_id_map
Args:
sentence (Dict): sentence: {'utt': ([char], [int])}
output_path (str or path):path to save phone_id_map
'''
phn_token = set()
for utt in sentence:
......@@ -112,14 +103,10 @@ def get_phones_tones(sentence,
dataset="baker"):
'''
get phone set and tone set from training data and save it
Parameters
----------
sentence : Dict
sentence: {'utt': ([char], [int])}
phones_output_path : str or path
path to save phone_id_map
tones_output_path : str or path
path to save tone_id_map
Args:
sentence (Dict): sentence: {'utt': ([char], [int])}
phones_output_path (str or path): path to save phone_id_map
tones_output_path (str or path): path to save tone_id_map
'''
phn_token = set()
tone_token = set()
......@@ -162,14 +149,10 @@ def get_spk_id_map(speaker_set, output_path):
def compare_duration_and_mel_length(sentences, utt, mel):
'''
check duration error, correct sentences[utt] if possible, else pop sentences[utt]
Parameters
----------
sentences : Dict
sentences[utt] = [phones_list ,durations_list]
utt : str
utt_id
mel : np.ndarry
features (num_frames, n_mels)
Args:
sentences (Dict): sentences[utt] = [phones_list ,durations_list]
utt (str): utt_id
mel (np.ndarry): features (num_frames, n_mels)
'''
if utt in sentences:
......
paddlespeech/t2s/datasets/vocoder_batch_fn.py
浏览文件 @ 9699c007
......@@ -29,15 +29,11 @@ class Clip(object):
hop_size=256,
aux_context_window=0, ):
"""Initialize customized collater for DataLoader.
Args:
Parameters
----------
batch_max_steps : int
The maximum length of input signal in batch.
hop_size : int
Hop size of auxiliary features.
aux_context_window : int
Context window size for auxiliary feature conv.
batch_max_steps (int): The maximum length of input signal in batch.
hop_size (int): Hop size of auxiliary features.
aux_context_window (int): Context window size for auxiliary feature conv.
"""
if batch_max_steps % hop_size != 0:
......@@ -56,18 +52,15 @@ class Clip(object):
def __call__(self, batch):
"""Convert into batch tensors.
Parameters
----------
batch : list
list of tuple of the pair of audio and features. Audio shape (T, ), features shape(T', C).
Args:
batch (list): list of tuple of the pair of audio and features. Audio shape (T, ), features shape(T', C).
Returns
----------
Tensor
Auxiliary feature batch (B, C, T'), where
T = (T' - 2 * aux_context_window) * hop_size.
Tensor
Target signal batch (B, 1, T).
Returns:
Tensor:
Auxiliary feature batch (B, C, T'), where
T = (T' - 2 * aux_context_window) * hop_size.
Tensor:
Target signal batch (B, 1, T).
"""
# check length
......@@ -104,11 +97,10 @@ class Clip(object):
def _adjust_length(self, x, c):
"""Adjust the audio and feature lengths.
Note
-------
Basically we assume that the length of x and c are adjusted
through preprocessing stage, but if we use other library processed
features, this process will be needed.
Note:
Basically we assume that the length of x and c are adjusted
through preprocessing stage, but if we use other library processed
features, this process will be needed.
"""
if len(x) < c.shape[0] * self.hop_size:
......@@ -162,22 +154,14 @@ class WaveRNNClip(Clip):
# voc_pad = 2 this will pad the input so that the resnet can 'see' wider than input length
# max_offsets = n_frames - 2 - (mel_win + 2 * hp.voc_pad) = n_frames - 15
"""Convert into batch tensors.
Parameters
----------
batch : list
list of tuple of the pair of audio and features.
Audio shape (T, ), features shape(T', C).
Returns
----------
Tensor
Input signal batch (B, 1, T).
Tensor
Target signal batch (B, 1, T).
Tensor
Auxiliary feature batch (B, C, T'), where
T = (T' - 2 * aux_context_window) * hop_size.
Args:
batch (list): list of tuple of the pair of audio and features. Audio shape (T, ), features shape(T', C).
Returns:
Tensor: Input signal batch (B, 1, T).
Tensor: Target signal batch (B, 1, T).
Tensor: Auxiliary feature batch (B, C, T'),
where T = (T' - 2 * aux_context_window) * hop_size.
"""
# check length
......
paddlespeech/t2s/exps/transformer_tts/preprocess.py
浏览文件 @ 9699c007
......@@ -31,15 +31,12 @@ from paddlespeech.t2s.frontend import English
def get_lj_sentences(file_name, frontend):
'''
read MFA duration.txt
Parameters
----------
file_name : str or Path
Returns
----------
Dict
sentence: {'utt': ([char], [int])}
'''read MFA duration.txt
Args:
file_name (str or Path)
Returns:
Dict: sentence: {'utt': ([char], [int])}
'''
f = open(file_name, 'r')
sentence = {}
......@@ -59,14 +56,11 @@ def get_lj_sentences(file_name, frontend):
def get_input_token(sentence, output_path):
'''
get phone set from training data and save it
Parameters
----------
sentence : Dict
sentence: {'utt': ([char], str)}
output_path : str or path
path to save phone_id_map
'''get phone set from training data and save it
Args:
sentence (Dict): sentence: {'utt': ([char], str)}
output_path (str or path): path to save phone_id_map
'''
phn_token = set()
for utt in sentence:
......
paddlespeech/t2s/frontend/arpabet.py
浏览文件 @ 9699c007
......@@ -133,16 +133,11 @@ class ARPABET(Phonetics):
def phoneticize(self, sentence, add_start_end=False):
""" Normalize the input text sequence and convert it into pronunciation sequence.
Args:
sentence (str): The input text sequence.
Parameters
-----------
sentence: str
The input text sequence.
Returns
----------
List[str]
The list of pronunciation sequence.
Returns:
List[str]: The list of pronunciation sequence.
"""
phonemes = [
self._remove_vowels(item) for item in self.backend(sentence)
......@@ -156,16 +151,12 @@ class ARPABET(Phonetics):
def numericalize(self, phonemes):
""" Convert pronunciation sequence into pronunciation id sequence.
Parameters
-----------
phonemes: List[str]
The list of pronunciation sequence.
Args:
phonemes (List[str]): The list of pronunciation sequence.
Returns
----------
List[int]
The list of pronunciation id sequence.
Returns:
List[int]: The list of pronunciation id sequence.
"""
ids = [self.vocab.lookup(item) for item in phonemes]
return ids
......@@ -173,30 +164,23 @@ class ARPABET(Phonetics):
def reverse(self, ids):
""" Reverse the list of pronunciation id sequence to a list of pronunciation sequence.
Parameters
-----------
ids: List[int]
The list of pronunciation id sequence.
Args:
ids( List[int]): The list of pronunciation id sequence.
Returns
----------
List[str]
The list of pronunciation sequence.
Returns:
List[str]:
The list of pronunciation sequence.
"""
return [self.vocab.reverse(i) for i in ids]
def __call__(self, sentence, add_start_end=False):
""" Convert the input text sequence into pronunciation id sequence.
Parameters
-----------
sentence: str
The input text sequence.
Args:
sentence (str): The input text sequence.
Returns
----------
List[str]
The list of pronunciation id sequence.
Returns:
List[str]: The list of pronunciation id sequence.
"""
return self.numericalize(
self.phoneticize(sentence, add_start_end=add_start_end))
......@@ -229,15 +213,11 @@ class ARPABETWithStress(Phonetics):
def phoneticize(self, sentence, add_start_end=False):
""" Normalize the input text sequence and convert it into pronunciation sequence.
Parameters
-----------
sentence: str
The input text sequence.
Args:
sentence (str): The input text sequence.
Returns
----------
List[str]
The list of pronunciation sequence.
Returns:
List[str]: The list of pronunciation sequence.
"""
phonemes = self.backend(sentence)
if add_start_end:
......@@ -249,47 +229,33 @@ class ARPABETWithStress(Phonetics):
def numericalize(self, phonemes):
""" Convert pronunciation sequence into pronunciation id sequence.
Parameters
-----------
phonemes: List[str]
The list of pronunciation sequence.
Args:
phonemes (List[str]): The list of pronunciation sequence.
Returns
----------
List[int]
The list of pronunciation id sequence.
Returns:
List[int]: The list of pronunciation id sequence.
"""
ids = [self.vocab.lookup(item) for item in phonemes]
return ids
def reverse(self, ids):
""" Reverse the list of pronunciation id sequence to a list of pronunciation sequence.
Parameters
-----------
ids: List[int]
The list of pronunciation id sequence.
Args:
ids (List[int]): The list of pronunciation id sequence.
Returns
----------
List[str]
The list of pronunciation sequence.
Returns:
List[str]: The list of pronunciation sequence.
"""
return [self.vocab.reverse(i) for i in ids]
def __call__(self, sentence, add_start_end=False):
""" Convert the input text sequence into pronunciation id sequence.
Args:
sentence (str): The input text sequence.
Parameters
-----------
sentence: str
The input text sequence.
Returns
----------
List[str]
The list of pronunciation id sequence.
Returns:
List[str]: The list of pronunciation id sequence.
"""
return self.numericalize(
self.phoneticize(sentence, add_start_end=add_start_end))
......
paddlespeech/t2s/frontend/phonectic.py
浏览文件 @ 9699c007
......@@ -65,14 +65,10 @@ class English(Phonetics):
def phoneticize(self, sentence):
""" Normalize the input text sequence and convert it into pronunciation sequence.
Parameters
-----------
sentence: str
The input text sequence.
Returns
----------
List[str]
The list of pronunciation sequence.
Args:
sentence (str): The input text sequence.
Returns:
List[str]: The list of pronunciation sequence.
"""
start = self.vocab.start_symbol
end = self.vocab.end_symbol
......@@ -123,14 +119,10 @@ class English(Phonetics):
def numericalize(self, phonemes):
""" Convert pronunciation sequence into pronunciation id sequence.
Parameters
-----------
phonemes: List[str]
The list of pronunciation sequence.
Returns
----------
List[int]
The list of pronunciation id sequence.
Args:
phonemes (List[str]): The list of pronunciation sequence.
Returns:
List[int]: The list of pronunciation id sequence.
"""
ids = [
self.vocab.lookup(item) for item in phonemes
......@@ -140,27 +132,19 @@ class English(Phonetics):
def reverse(self, ids):
""" Reverse the list of pronunciation id sequence to a list of pronunciation sequence.
Parameters
-----------
ids: List[int]
The list of pronunciation id sequence.
Returns
----------
List[str]
The list of pronunciation sequence.
Args:
ids (List[int]): The list of pronunciation id sequence.
Returns:
List[str]: The list of pronunciation sequence.
"""
return [self.vocab.reverse(i) for i in ids]
def __call__(self, sentence):
""" Convert the input text sequence into pronunciation id sequence.
Parameters
-----------
sentence: str
The input text sequence.
Returns
----------
List[str]
The list of pronunciation id sequence.
Args:
sentence(str): The input text sequence.
Returns:
List[str]: The list of pronunciation id sequence.
"""
return self.numericalize(self.phoneticize(sentence))
......@@ -183,28 +167,21 @@ class EnglishCharacter(Phonetics):
def phoneticize(self, sentence):
""" Normalize the input text sequence.
Parameters
-----------
sentence: str
The input text sequence.
Returns
----------
str
A text sequence after normalize.
Args:
sentence(str): The input text sequence.
Returns:
str: A text sequence after normalize.
"""
words = normalize(sentence)
return words
def numericalize(self, sentence):
""" Convert a text sequence into ids.
Parameters
-----------
sentence: str
The input text sequence.
Returns
----------
List[int]
List of a character id sequence.
Args:
sentence (str): The input text sequence.
Returns:
List[int]:
List of a character id sequence.
"""
ids = [
self.vocab.lookup(item) for item in sentence
......@@ -214,27 +191,19 @@ class EnglishCharacter(Phonetics):
def reverse(self, ids):
""" Convert a character id sequence into text.
Parameters
-----------
ids: List[int]
List of a character id sequence.
Returns
----------
str
The input text sequence.
Args:
ids (List[int]): List of a character id sequence.
Returns:
str: The input text sequence.
"""
return [self.vocab.reverse(i) for i in ids]
def __call__(self, sentence):
""" Normalize the input text sequence and convert it into character id sequence.
Parameters
-----------
sentence: str
The input text sequence.
Returns
----------
List[int]
List of a character id sequence.
Args:
sentence (str): The input text sequence.
Returns:
List[int]: List of a character id sequence.
"""
return self.numericalize(self.phoneticize(sentence))
......@@ -264,14 +233,10 @@ class Chinese(Phonetics):
def phoneticize(self, sentence):
""" Normalize the input text sequence and convert it into pronunciation sequence.
Parameters
-----------
sentence: str
The input text sequence.
Returns
----------
List[str]
The list of pronunciation sequence.
Args:
sentence(str): The input text sequence.
Returns:
List[str]: The list of pronunciation sequence.
"""
# simplified = self.opencc_backend.convert(sentence)
simplified = sentence
......@@ -296,28 +261,20 @@ class Chinese(Phonetics):
def numericalize(self, phonemes):
""" Convert pronunciation sequence into pronunciation id sequence.
Parameters
-----------
phonemes: List[str]
The list of pronunciation sequence.
Returns
----------
List[int]
The list of pronunciation id sequence.
Args:
phonemes(List[str]): The list of pronunciation sequence.
Returns:
List[int]: The list of pronunciation id sequence.
"""
ids = [self.vocab.lookup(item) for item in phonemes]
return ids
def __call__(self, sentence):
""" Convert the input text sequence into pronunciation id sequence.
Parameters
-----------
sentence: str
The input text sequence.
Returns
----------
List[str]
The list of pronunciation id sequence.
Args:
sentence (str): The input text sequence.
Returns:
List[str]: The list of pronunciation id sequence.
"""
return self.numericalize(self.phoneticize(sentence))
......@@ -329,13 +286,9 @@ class Chinese(Phonetics):
def reverse(self, ids):
""" Reverse the list of pronunciation id sequence to a list of pronunciation sequence.
Parameters
-----------
ids: List[int]
The list of pronunciation id sequence.
Returns
----------
List[str]
The list of pronunciation sequence.
Args:
ids (List[int]): The list of pronunciation id sequence.
Returns:
List[str]: The list of pronunciation sequence.
"""
return [self.vocab.reverse(i) for i in ids]
paddlespeech/t2s/frontend/vocab.py
浏览文件 @ 9699c007
......@@ -20,22 +20,12 @@ __all__ = ["Vocab"]
class Vocab(object):
""" Vocabulary.
Parameters
-----------
symbols: Iterable[str]
Common symbols.
padding_symbol: str, optional
Symbol for pad. Defaults to "<pad>".
unk_symbol: str, optional
Symbol for unknow. Defaults to "<unk>"
start_symbol: str, optional
Symbol for start. Defaults to "<s>"
end_symbol: str, optional
Symbol for end. Defaults to "</s>"
Args:
symbols (Iterable[str]): Common symbols.
padding_symbol (str, optional): Symbol for pad. Defaults to "<pad>".
unk_symbol (str, optional): Symbol for unknow. Defaults to "<unk>"
start_symbol (str, optional): Symbol for start. Defaults to "<s>"
end_symbol (str, optional): Symbol for end. Defaults to "</s>"
"""
def __init__(self,
......
paddlespeech/t2s/frontend/zh_normalization/chronology.py
浏览文件 @ 9699c007
......@@ -44,12 +44,10 @@ RE_TIME_RANGE = re.compile(r'([0-1]?[0-9]|2[0-3])'
def replace_time(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
is_range = len(match.groups()) > 5
......@@ -87,12 +85,10 @@ RE_DATE = re.compile(r'(\d{4}|\d{2})年'
def replace_date(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
year = match.group(1)
month = match.group(3)
......@@ -114,12 +110,10 @@ RE_DATE2 = re.compile(
def replace_date2(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
year = match.group(1)
month = match.group(3)
......
paddlespeech/t2s/frontend/zh_normalization/num.py
浏览文件 @ 9699c007
......@@ -36,12 +36,10 @@ RE_FRAC = re.compile(r'(-?)(\d+)/(\d+)')
def replace_frac(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
sign = match.group(1)
nominator = match.group(2)
......@@ -59,12 +57,10 @@ RE_PERCENTAGE = re.compile(r'(-?)(\d+(\.\d+)?)%')
def replace_percentage(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
sign = match.group(1)
percent = match.group(2)
......@@ -81,12 +77,10 @@ RE_INTEGER = re.compile(r'(-)' r'(\d+)')
def replace_negative_num(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
sign = match.group(1)
number = match.group(2)
......@@ -103,12 +97,10 @@ RE_DEFAULT_NUM = re.compile(r'\d{3}\d*')
def replace_default_num(match):
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
number = match.group(0)
return verbalize_digit(number)
......@@ -124,12 +116,10 @@ RE_NUMBER = re.compile(r'(-?)((\d+)(\.\d+)?)' r'|(\.(\d+))')
def replace_positive_quantifier(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
number = match.group(1)
match_2 = match.group(2)
......@@ -142,12 +132,10 @@ def replace_positive_quantifier(match) -> str:
def replace_number(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
sign = match.group(1)
number = match.group(2)
......@@ -169,12 +157,10 @@ RE_RANGE = re.compile(
def replace_range(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
first, second = match.group(1), match.group(8)
first = RE_NUMBER.sub(replace_number, first)
......
paddlespeech/t2s/frontend/zh_normalization/phonecode.py
浏览文件 @ 9699c007
......@@ -45,23 +45,19 @@ def phone2str(phone_string: str, mobile=True) -> str:
def replace_phone(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
return phone2str(match.group(0), mobile=False)
def replace_mobile(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
return phone2str(match.group(0))
paddlespeech/t2s/frontend/zh_normalization/quantifier.py
浏览文件 @ 9699c007
......@@ -22,12 +22,10 @@ RE_TEMPERATURE = re.compile(r'(-?)(\d+(\.\d+)?)(°C|℃|度|摄氏度)')
def replace_temperature(match) -> str:
"""
Parameters
----------
match : re.Match
Returns
----------
str
Args:
match (re.Match)
Returns:
str
"""
sign = match.group(1)
temperature = match.group(2)
......
paddlespeech/t2s/frontend/zh_normalization/text_normlization.py
浏览文件 @ 9699c007
......@@ -55,14 +55,10 @@ class TextNormalizer():
def _split(self, text: str, lang="zh") -> List[str]:
"""Split long text into sentences with sentence-splitting punctuations.
Parameters
----------
text : str
The input text.
Returns
-------
List[str]
Sentences.
Args:
text (str): The input text.
Returns:
List[str]: Sentences.
"""
# Only for pure Chinese here
if lang == "zh":
......
paddlespeech/t2s/models/hifigan/hifigan.py
浏览文件 @ 9699c007
......@@ -37,35 +37,21 @@ class HiFiGANGenerator(nn.Layer):
use_weight_norm: bool=True,
init_type: str="xavier_uniform", ):
"""Initialize HiFiGANGenerator module.
Parameters
----------
in_channels : int
Number of input channels.
out_channels : int
Number of output channels.
channels : int
Number of hidden representation channels.
kernel_size : int
Kernel size of initial and final conv layer.
upsample_scales : list
List of upsampling scales.
upsample_kernel_sizes : list
List of kernel sizes for upsampling layers.
resblock_kernel_sizes : list
List of kernel sizes for residual blocks.
resblock_dilations : list
List of dilation list for residual blocks.
use_additional_convs : bool
Whether to use additional conv layers in residual blocks.
bias : bool
Whether to add bias parameter in convolution layers.
nonlinear_activation : str
Activation function module name.
nonlinear_activation_params : dict
Hyperparameters for activation function.
use_weight_norm : bool
Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
channels (int): Number of hidden representation channels.
kernel_size (int): Kernel size of initial and final conv layer.
upsample_scales (list): List of upsampling scales.
upsample_kernel_sizes (list): List of kernel sizes for upsampling layers.
resblock_kernel_sizes (list): List of kernel sizes for residual blocks.
resblock_dilations (list): List of dilation list for residual blocks.
use_additional_convs (bool): Whether to use additional conv layers in residual blocks.
bias (bool): Whether to add bias parameter in convolution layers.
nonlinear_activation (str): Activation function module name.
nonlinear_activation_params (dict): Hyperparameters for activation function.
use_weight_norm (bool): Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
"""
super().__init__()
......@@ -134,14 +120,11 @@ class HiFiGANGenerator(nn.Layer):
def forward(self, c):
"""Calculate forward propagation.
Parameters
----------
c : Tensor
Input tensor (B, in_channels, T).
Returns
----------
Tensor
Output tensor (B, out_channels, T).
Args:
c (Tensor): Input tensor (B, in_channels, T).
Returns:
Tensor: Output tensor (B, out_channels, T).
"""
c = self.input_conv(c)
for i in range(self.num_upsamples):
......@@ -196,15 +179,12 @@ class HiFiGANGenerator(nn.Layer):
def inference(self, c):
"""Perform inference.
Parameters
----------
c : Tensor
Input tensor (T, in_channels).
normalize_before (bool): Whether to perform normalization.
Returns
----------
Tensor
Output tensor (T ** prod(upsample_scales), out_channels).
Args:
c (Tensor): Input tensor (T, in_channels).
normalize_before (bool): Whether to perform normalization.
Returns:
Tensor:
Output tensor (T ** prod(upsample_scales), out_channels).
"""
c = self.forward(c.transpose([1, 0]).unsqueeze(0))
return c.squeeze(0).transpose([1, 0])
......@@ -229,36 +209,23 @@ class HiFiGANPeriodDiscriminator(nn.Layer):
use_spectral_norm: bool=False,
init_type: str="xavier_uniform", ):
"""Initialize HiFiGANPeriodDiscriminator module.
Parameters
----------
in_channels : int
Number of input channels.
out_channels : int
Number of output channels.
period : int
Period.
kernel_sizes : list
Kernel sizes of initial conv layers and the final conv layer.
channels : int
Number of initial channels.
downsample_scales : list
List of downsampling scales.
max_downsample_channels : int
Number of maximum downsampling channels.
use_additional_convs : bool
Whether to use additional conv layers in residual blocks.
bias : bool
Whether to add bias parameter in convolution layers.
nonlinear_activation : str
Activation function module name.
nonlinear_activation_params : dict
Hyperparameters for activation function.
use_weight_norm : bool
Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_spectral_norm : bool
Whether to use spectral norm.
If set to true, it will be applied to all of the conv layers.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
period (int): Period.
kernel_sizes (list): Kernel sizes of initial conv layers and the final conv layer.
channels (int): Number of initial channels.
downsample_scales (list): List of downsampling scales.
max_downsample_channels (int): Number of maximum downsampling channels.
use_additional_convs (bool): Whether to use additional conv layers in residual blocks.
bias (bool): Whether to add bias parameter in convolution layers.
nonlinear_activation (str): Activation function module name.
nonlinear_activation_params (dict): Hyperparameters for activation function.
use_weight_norm (bool): Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_spectral_norm (bool): Whether to use spectral norm.
If set to true, it will be applied to all of the conv layers.
"""
super().__init__()
......@@ -307,14 +274,11 @@ class HiFiGANPeriodDiscriminator(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
c : Tensor
Input tensor (B, in_channels, T).
Returns
----------
list
List of each layer's tensors.
Args:
c (Tensor): Input tensor (B, in_channels, T).
Returns:
list: List of each layer's tensors.
"""
# transform 1d to 2d -> (B, C, T/P, P)
b, c, t = paddle.shape(x)
......@@ -379,13 +343,11 @@ class HiFiGANMultiPeriodDiscriminator(nn.Layer):
},
init_type: str="xavier_uniform", ):
"""Initialize HiFiGANMultiPeriodDiscriminator module.
Parameters
----------
periods : list
List of periods.
discriminator_params : dict
Parameters for hifi-gan period discriminator module.
The period parameter will be overwritten.
Args:
periods (list): List of periods.
discriminator_params (dict): Parameters for hifi-gan period discriminator module.
The period parameter will be overwritten.
"""
super().__init__()
# initialize parameters
......@@ -399,14 +361,11 @@ class HiFiGANMultiPeriodDiscriminator(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input noise signal (B, 1, T).
Returns
----------
List
List of list of each discriminator outputs, which consists of each layer output tensors.
Args:
x (Tensor): Input noise signal (B, 1, T).
Returns:
List: List of list of each discriminator outputs, which consists of each layer output tensors.
"""
outs = []
for f in self.discriminators:
......@@ -434,33 +393,22 @@ class HiFiGANScaleDiscriminator(nn.Layer):
use_spectral_norm: bool=False,
init_type: str="xavier_uniform", ):
"""Initilize HiFiGAN scale discriminator module.
Parameters
----------
in_channels : int
Number of input channels.
out_channels : int
Number of output channels.
kernel_sizes : list
List of four kernel sizes. The first will be used for the first conv layer,
and the second is for downsampling part, and the remaining two are for output layers.
channels : int
Initial number of channels for conv layer.
max_downsample_channels : int
Maximum number of channels for downsampling layers.
bias : bool
Whether to add bias parameter in convolution layers.
downsample_scales : list
List of downsampling scales.
nonlinear_activation : str
Activation function module name.
nonlinear_activation_params : dict
Hyperparameters for activation function.
use_weight_norm : bool
Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_spectral_norm : bool
Whether to use spectral norm.
If set to true, it will be applied to all of the conv layers.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
kernel_sizes (list): List of four kernel sizes. The first will be used for the first conv layer,
and the second is for downsampling part, and the remaining two are for output layers.
channels (int): Initial number of channels for conv layer.
max_downsample_channels (int): Maximum number of channels for downsampling layers.
bias (bool): Whether to add bias parameter in convolution layers.
downsample_scales (list): List of downsampling scales.
nonlinear_activation (str): Activation function module name.
nonlinear_activation_params (dict): Hyperparameters for activation function.
use_weight_norm (bool): Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_spectral_norm (bool): Whether to use spectral norm.
If set to true, it will be applied to all of the conv layers.
"""
super().__init__()
......@@ -546,14 +494,11 @@ class HiFiGANScaleDiscriminator(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input noise signal (B, 1, T).
Returns
----------
List
List of output tensors of each layer.
Args:
x (Tensor): Input noise signal (B, 1, T).
Returns:
List: List of output tensors of each layer.
"""
outs = []
for f in self.layers:
......@@ -613,20 +558,14 @@ class HiFiGANMultiScaleDiscriminator(nn.Layer):
follow_official_norm: bool=False,
init_type: str="xavier_uniform", ):
"""Initilize HiFiGAN multi-scale discriminator module.
Parameters
----------
scales : int
Number of multi-scales.
downsample_pooling : str
Pooling module name for downsampling of the inputs.
downsample_pooling_params : dict
Parameters for the above pooling module.
discriminator_params : dict
Parameters for hifi-gan scale discriminator module.
follow_official_norm : bool
Whether to follow the norm setting of the official
implementaion. The first discriminator uses spectral norm and the other
discriminators use weight norm.
Args:
scales (int): Number of multi-scales.
downsample_pooling (str): Pooling module name for downsampling of the inputs.
downsample_pooling_params (dict): Parameters for the above pooling module.
discriminator_params (dict): Parameters for hifi-gan scale discriminator module.
follow_official_norm (bool): Whether to follow the norm setting of the official
implementaion. The first discriminator uses spectral norm and the other discriminators use weight norm.
"""
super().__init__()
......@@ -651,14 +590,11 @@ class HiFiGANMultiScaleDiscriminator(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input noise signal (B, 1, T).
Returns
----------
List
List of list of each discriminator outputs, which consists of each layer output tensors.
Args:
x (Tensor): Input noise signal (B, 1, T).
Returns:
List: List of list of each discriminator outputs, which consists of each layer output tensors.
"""
outs = []
for f in self.discriminators:
......@@ -715,24 +651,17 @@ class HiFiGANMultiScaleMultiPeriodDiscriminator(nn.Layer):
},
init_type: str="xavier_uniform", ):
"""Initilize HiFiGAN multi-scale + multi-period discriminator module.
Parameters
----------
scales : int
Number of multi-scales.
scale_downsample_pooling : str
Pooling module name for downsampling of the inputs.
scale_downsample_pooling_params : dict
Parameters for the above pooling module.
scale_discriminator_params : dict
Parameters for hifi-gan scale discriminator module.
follow_official_norm : bool): Whether to follow the norm setting of the official
implementaion. The first discriminator uses spectral norm and the other
discriminators use weight norm.
periods : list
List of periods.
period_discriminator_params : dict
Parameters for hifi-gan period discriminator module.
The period parameter will be overwritten.
Args:
scales (int): Number of multi-scales.
scale_downsample_pooling (str): Pooling module name for downsampling of the inputs.
scale_downsample_pooling_params (dict): Parameters for the above pooling module.
scale_discriminator_params (dict): Parameters for hifi-gan scale discriminator module.
follow_official_norm （bool): Whether to follow the norm setting of the official implementaion.
The first discriminator uses spectral norm and the other discriminators use weight norm.
periods (list): List of periods.
period_discriminator_params (dict): Parameters for hifi-gan period discriminator module.
The period parameter will be overwritten.
"""
super().__init__()
......@@ -751,16 +680,14 @@ class HiFiGANMultiScaleMultiPeriodDiscriminator(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input noise signal (B, 1, T).
Returns
----------
List:
List of list of each discriminator outputs,
which consists of each layer output tensors.
Multi scale and multi period ones are concatenated.
Args:
x (Tensor): Input noise signal (B, 1, T).
Returns:
List:
List of list of each discriminator outputs,
which consists of each layer output tensors.
Multi scale and multi period ones are concatenated.
"""
msd_outs = self.msd(x)
mpd_outs = self.mpd(x)
......
paddlespeech/t2s/models/melgan/melgan.py
浏览文件 @ 9699c007
......@@ -51,41 +51,26 @@ class MelGANGenerator(nn.Layer):
use_causal_conv: bool=False,
init_type: str="xavier_uniform", ):
"""Initialize MelGANGenerator module.
Parameters
----------
in_channels : int
Number of input channels.
out_channels : int
Number of output channels,
the number of sub-band is out_channels in multi-band melgan.
kernel_size : int
Kernel size of initial and final conv layer.
channels : int
Initial number of channels for conv layer.
bias : bool
Whether to add bias parameter in convolution layers.
upsample_scales : List[int]
List of upsampling scales.
stack_kernel_size : int
Kernel size of dilated conv layers in residual stack.
stacks : int
Number of stacks in a single residual stack.
nonlinear_activation : Optional[str], optional
Non linear activation in upsample network, by default None
nonlinear_activation_params : Dict[str, Any], optional
Parameters passed to the linear activation in the upsample network,
by default {}
pad : str
Padding function module name before dilated convolution layer.
pad_params : dict
Hyperparameters for padding function.
use_final_nonlinear_activation : nn.Layer
Activation function for the final layer.
use_weight_norm : bool
Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_causal_conv : bool
Whether to use causal convolution.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels,
the number of sub-band is out_channels in multi-band melgan.
kernel_size (int): Kernel size of initial and final conv layer.
channels (int): Initial number of channels for conv layer.
bias (bool): Whether to add bias parameter in convolution layers.
upsample_scales (List[int]): List of upsampling scales.
stack_kernel_size (int): Kernel size of dilated conv layers in residual stack.
stacks (int): Number of stacks in a single residual stack.
nonlinear_activation (Optional[str], optional): Non linear activation in upsample network, by default None
nonlinear_activation_params (Dict[str, Any], optional): Parameters passed to the linear activation in the upsample network,
by default {}
pad (str): Padding function module name before dilated convolution layer.
pad_params （dict): Hyperparameters for padding function.
use_final_nonlinear_activation (nn.Layer): Activation function for the final layer.
use_weight_norm (bool): Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_causal_conv (bool): Whether to use causal convolution.
"""
super().__init__()
......@@ -207,14 +192,11 @@ class MelGANGenerator(nn.Layer):
def forward(self, c):
"""Calculate forward propagation.
Parameters
----------
c : Tensor
Input tensor (B, in_channels, T).
Returns
----------
Tensor
Output tensor (B, out_channels, T ** prod(upsample_scales)).
Args:
c (Tensor): Input tensor (B, in_channels, T).
Returns:
Tensor: Output tensor (B, out_channels, T ** prod(upsample_scales)).
"""
out = self.melgan(c)
return out
......@@ -260,14 +242,11 @@ class MelGANGenerator(nn.Layer):
def inference(self, c):
"""Perform inference.
Parameters
----------
c : Union[Tensor, ndarray]
Input tensor (T, in_channels).
Returns
----------
Tensor
Output tensor (out_channels*T ** prod(upsample_scales), 1).
Args:
c (Union[Tensor, ndarray]): Input tensor (T, in_channels).
Returns:
Tensor: Output tensor (out_channels*T ** prod(upsample_scales), 1).
"""
# pseudo batch
c = c.transpose([1, 0]).unsqueeze(0)
......@@ -298,33 +277,22 @@ class MelGANDiscriminator(nn.Layer):
pad_params: Dict[str, Any]={"mode": "reflect"},
init_type: str="xavier_uniform", ):
"""Initilize MelGAN discriminator module.
Parameters
----------
in_channels : int
Number of input channels.
out_channels : int
Number of output channels.
kernel_sizes : List[int]
List of two kernel sizes. The prod will be used for the first conv layer,
and the first and the second kernel sizes will be used for the last two layers.
For example if kernel_sizes = [5, 3], the first layer kernel size will be 5 * 3 = 15,
the last two layers' kernel size will be 5 and 3, respectively.
channels : int
Initial number of channels for conv layer.
max_downsample_channels : int
Maximum number of channels for downsampling layers.
bias : bool
Whether to add bias parameter in convolution layers.
downsample_scales : List[int]
List of downsampling scales.
nonlinear_activation : str
Activation function module name.
nonlinear_activation_params : dict
Hyperparameters for activation function.
pad : str
Padding function module name before dilated convolution layer.
pad_params : dict
Hyperparameters for padding function.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
kernel_sizes (List[int]): List of two kernel sizes. The prod will be used for the first conv layer,
and the first and the second kernel sizes will be used for the last two layers.
For example if kernel_sizes = [5, 3], the first layer kernel size will be 5 * 3 = 15,
the last two layers' kernel size will be 5 and 3, respectively.
channels (int): Initial number of channels for conv layer.
max_downsample_channels (int): Maximum number of channels for downsampling layers.
bias (bool): Whether to add bias parameter in convolution layers.
downsample_scales (List[int]): List of downsampling scales.
nonlinear_activation (str): Activation function module name.
nonlinear_activation_params (dict): Hyperparameters for activation function.
pad (str): Padding function module name before dilated convolution layer.
pad_params (dict): Hyperparameters for padding function.
"""
super().__init__()
......@@ -395,14 +363,10 @@ class MelGANDiscriminator(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input noise signal (B, 1, T).
Returns
----------
List
List of output tensors of each layer (for feat_match_loss).
Args:
x (Tensor): Input noise signal (B, 1, T).
Returns:
List: List of output tensors of each layer (for feat_match_loss).
"""
outs = []
for f in self.layers:
......@@ -440,39 +404,24 @@ class MelGANMultiScaleDiscriminator(nn.Layer):
use_weight_norm: bool=True,
init_type: str="xavier_uniform", ):
"""Initilize MelGAN multi-scale discriminator module.
Parameters
----------
in_channels : int
Number of input channels.
out_channels : int
Number of output channels.
scales : int
Number of multi-scales.
downsample_pooling : str
Pooling module name for downsampling of the inputs.
downsample_pooling_params : dict
Parameters for the above pooling module.
kernel_sizes : List[int]
List of two kernel sizes. The sum will be used for the first conv layer,
and the first and the second kernel sizes will be used for the last two layers.
channels : int
Initial number of channels for conv layer.
max_downsample_channels : int
Maximum number of channels for downsampling layers.
bias : bool
Whether to add bias parameter in convolution layers.
downsample_scales : List[int]
List of downsampling scales.
nonlinear_activation : str
Activation function module name.
nonlinear_activation_params : dict
Hyperparameters for activation function.
pad : str
Padding function module name before dilated convolution layer.
pad_params : dict
Hyperparameters for padding function.
use_causal_conv : bool
Whether to use causal convolution.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
scales (int): Number of multi-scales.
downsample_pooling (str): Pooling module name for downsampling of the inputs.
downsample_pooling_params (dict): Parameters for the above pooling module.
kernel_sizes (List[int]): List of two kernel sizes. The sum will be used for the first conv layer,
and the first and the second kernel sizes will be used for the last two layers.
channels (int): Initial number of channels for conv layer.
max_downsample_channels (int): Maximum number of channels for downsampling layers.
bias (bool): Whether to add bias parameter in convolution layers.
downsample_scales (List[int]): List of downsampling scales.
nonlinear_activation (str): Activation function module name.
nonlinear_activation_params (dict): Hyperparameters for activation function.
pad (str): Padding function module name before dilated convolution layer.
pad_params (dict): Hyperparameters for padding function.
use_causal_conv (bool): Whether to use causal convolution.
"""
super().__init__()
......@@ -514,14 +463,10 @@ class MelGANMultiScaleDiscriminator(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input noise signal (B, 1, T).
Returns
----------
List
List of list of each discriminator outputs, which consists of each layer output tensors.
Args:
x (Tensor): Input noise signal (B, 1, T).
Returns:
List: List of list of each discriminator outputs, which consists of each layer output tensors.
"""
outs = []
for f in self.discriminators:
......
paddlespeech/t2s/models/melgan/style_melgan.py
浏览文件 @ 9699c007
......@@ -52,37 +52,23 @@ class StyleMelGANGenerator(nn.Layer):
use_weight_norm: bool=True,
init_type: str="xavier_uniform", ):
"""Initilize Style MelGAN generator.
Parameters
----------
in_channels : int
Number of input noise channels.
aux_channels : int
Number of auxiliary input channels.
channels : int
Number of channels for conv layer.
out_channels : int
Number of output channels.
kernel_size : int
Kernel size of conv layers.
dilation : int
Dilation factor for conv layers.
bias : bool
Whether to add bias parameter in convolution layers.
noise_upsample_scales : list
List of noise upsampling scales.
noise_upsample_activation : str
Activation function module name for noise upsampling.
noise_upsample_activation_params : dict
Hyperparameters for the above activation function.
upsample_scales : list
List of upsampling scales.
upsample_mode : str
Upsampling mode in TADE layer.
gated_function : str
Gated function in TADEResBlock ("softmax" or "sigmoid").
use_weight_norm : bool
Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
Args:
in_channels (int): Number of input noise channels.
aux_channels (int): Number of auxiliary input channels.
channels (int): Number of channels for conv layer.
out_channels (int): Number of output channels.
kernel_size (int): Kernel size of conv layers.
dilation (int): Dilation factor for conv layers.
bias (bool): Whether to add bias parameter in convolution layers.
noise_upsample_scales (list): List of noise upsampling scales.
noise_upsample_activation (str): Activation function module name for noise upsampling.
noise_upsample_activation_params (dict): Hyperparameters for the above activation function.
upsample_scales (list): List of upsampling scales.
upsample_mode (str): Upsampling mode in TADE layer.
gated_function (str): Gated function in TADEResBlock ("softmax" or "sigmoid").
use_weight_norm (bool): Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
"""
super().__init__()
......@@ -147,16 +133,12 @@ class StyleMelGANGenerator(nn.Layer):
def forward(self, c, z=None):
"""Calculate forward propagation.
Parameters
----------
c : Tensor
Auxiliary input tensor (B, channels, T).
z : Tensor
Input noise tensor (B, in_channels, 1).
Returns
----------
Tensor
Output tensor (B, out_channels, T ** prod(upsample_scales)).
Args:
c (Tensor): Auxiliary input tensor (B, channels, T).
z (Tensor): Input noise tensor (B, in_channels, 1).
Returns:
Tensor: Output tensor (B, out_channels, T ** prod(upsample_scales)).
"""
# batch_max_steps(24000) == noise_upsample_factor(80) * upsample_factor(300)
if z is None:
......@@ -211,14 +193,10 @@ class StyleMelGANGenerator(nn.Layer):
def inference(self, c):
"""Perform inference.
Parameters
----------
c : Tensor
Input tensor (T, in_channels).
Returns
----------
Tensor
Output tensor (T ** prod(upsample_scales), out_channels).
Args:
c (Tensor): Input tensor (T, in_channels).
Returns:
Tensor: Output tensor (T ** prod(upsample_scales), out_channels).
"""
# (1, in_channels, T)
c = c.transpose([1, 0]).unsqueeze(0)
......@@ -278,18 +256,13 @@ class StyleMelGANDiscriminator(nn.Layer):
use_weight_norm: bool=True,
init_type: str="xavier_uniform", ):
"""Initilize Style MelGAN discriminator.
Parameters
----------
repeats : int
Number of repititons to apply RWD.
window_sizes : list
List of random window sizes.
pqmf_params : list
List of list of Parameters for PQMF modules
discriminator_params : dict
Parameters for base discriminator module.
use_weight_nom : bool
Whether to apply weight normalization.
Args:
repeats (int): Number of repititons to apply RWD.
window_sizes (list): List of random window sizes.
pqmf_params (list): List of list of Parameters for PQMF modules
discriminator_params (dict): Parameters for base discriminator module.
use_weight_nom (bool): Whether to apply weight normalization.
"""
super().__init__()
......@@ -325,15 +298,11 @@ class StyleMelGANDiscriminator(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input tensor (B, 1, T).
Returns
----------
List
List of discriminator outputs, #items in the list will be
equal to repeats * #discriminators.
Args:
x (Tensor): Input tensor (B, 1, T).
Returns:
List: List of discriminator outputs, #items in the list will be
equal to repeats * #discriminators.
"""
outs = []
for _ in range(self.repeats):
......
paddlespeech/t2s/models/parallel_wavegan/parallel_wavegan.py
浏览文件 @ 9699c007
......@@ -31,51 +31,30 @@ from paddlespeech.t2s.modules.upsample import ConvInUpsampleNet
class PWGGenerator(nn.Layer):
"""Wave Generator for Parallel WaveGAN
Parameters
----------
in_channels : int, optional
Number of channels of the input waveform, by default 1
out_channels : int, optional
Number of channels of the output waveform, by default 1
kernel_size : int, optional
Kernel size of the residual blocks inside, by default 3
layers : int, optional
Number of residual blocks inside, by default 30
stacks : int, optional
The number of groups to split the residual blocks into, by default 3
Within each group, the dilation of the residual block grows
exponentially.
residual_channels : int, optional
Residual channel of the residual blocks, by default 64
gate_channels : int, optional
Gate channel of the residual blocks, by default 128
skip_channels : int, optional
Skip channel of the residual blocks, by default 64
aux_channels : int, optional
Auxiliary channel of the residual blocks, by default 80
aux_context_window : int, optional
The context window size of the first convolution applied to the
auxiliary input, by default 2
dropout : float, optional
Dropout of the residual blocks, by default 0.
bias : bool, optional
Whether to use bias in residual blocks, by default True
use_weight_norm : bool, optional
Whether to use weight norm in all convolutions, by default True
use_causal_conv : bool, optional
Whether to use causal padding in the upsample network and residual
blocks, by default False
upsample_scales : List[int], optional
Upsample scales of the upsample network, by default [4, 4, 4, 4]
nonlinear_activation : Optional[str], optional
Non linear activation in upsample network, by default None
nonlinear_activation_params : Dict[str, Any], optional
Parameters passed to the linear activation in the upsample network,
by default {}
interpolate_mode : str, optional
Interpolation mode of the upsample network, by default "nearest"
freq_axis_kernel_size : int, optional
Kernel size along the frequency axis of the upsample network, by default 1
Args:
in_channels (int, optional): Number of channels of the input waveform, by default 1
out_channels (int, optional): Number of channels of the output waveform, by default 1
kernel_size (int, optional): Kernel size of the residual blocks inside, by default 3
layers (int, optional): Number of residual blocks inside, by default 30
stacks (int, optional): The number of groups to split the residual blocks into, by default 3
Within each group, the dilation of the residual block grows exponentially.
residual_channels (int, optional): Residual channel of the residual blocks, by default 64
gate_channels (int, optional): Gate channel of the residual blocks, by default 128
skip_channels (int, optional): Skip channel of the residual blocks, by default 64
aux_channels (int, optional): Auxiliary channel of the residual blocks, by default 80
aux_context_window (int, optional): The context window size of the first convolution applied to the
auxiliary input, by default 2
dropout (float, optional): Dropout of the residual blocks, by default 0.
bias (bool, optional): Whether to use bias in residual blocks, by default True
use_weight_norm (bool, optional): Whether to use weight norm in all convolutions, by default True
use_causal_conv (bool, optional): Whether to use causal padding in the upsample network and residual
blocks, by default False
upsample_scales (List[int], optional): Upsample scales of the upsample network, by default [4, 4, 4, 4]
nonlinear_activation (Optional[str], optional): Non linear activation in upsample network, by default None
nonlinear_activation_params (Dict[str, Any], optional): Parameters passed to the linear activation in the upsample network,
by default {}
interpolate_mode (str, optional): Interpolation mode of the upsample network, by default "nearest"
freq_axis_kernel_size (int, optional): Kernel size along the frequency axis of the upsample network, by default 1
"""
def __init__(
......@@ -167,18 +146,13 @@ class PWGGenerator(nn.Layer):
def forward(self, x, c):
"""Generate waveform.
Parameters
----------
x : Tensor
Shape (N, C_in, T), The input waveform.
c : Tensor
Shape (N, C_aux, T'). The auxiliary input (e.g. spectrogram). It
Args:
x(Tensor): Shape (N, C_in, T), The input waveform.
c(Tensor): Shape (N, C_aux, T'). The auxiliary input (e.g. spectrogram). It
is upsampled to match the time resolution of the input.
Returns
-------
Tensor
Shape (N, C_out, T), the generated waveform.
Returns:
Tensor: Shape (N, C_out, T), the generated waveform.
"""
c = self.upsample_net(c)
assert c.shape[-1] == x.shape[-1]
......@@ -218,19 +192,14 @@ class PWGGenerator(nn.Layer):
self.apply(_remove_weight_norm)
def inference(self, c=None):
"""Waveform generation. This function is used for single instance
inference.
Parameters
----------
c : Tensor, optional
Shape (T', C_aux), the auxiliary input, by default None
x : Tensor, optional
Shape (T, C_in), the noise waveform, by default None
If not provided, a sample is drawn from a gaussian distribution.
Returns
-------
Tensor
Shape (T, C_out), the generated waveform
"""Waveform generation. This function is used for single instance inference.
Args:
c(Tensor, optional, optional): Shape (T', C_aux), the auxiliary input, by default None
x(Tensor, optional): Shape (T, C_in), the noise waveform, by default None
Returns:
Tensor: Shape (T, C_out), the generated waveform
"""
# when to static, can not input x, see https://github.com/PaddlePaddle/Parakeet/pull/132/files
x = paddle.randn(
......@@ -244,32 +213,21 @@ class PWGGenerator(nn.Layer):
class PWGDiscriminator(nn.Layer):
"""A convolutional discriminator for audio.
Parameters
----------
in_channels : int, optional
Number of channels of the input audio, by default 1
out_channels : int, optional
Output feature size, by default 1
kernel_size : int, optional
Kernel size of convolutional sublayers, by default 3
layers : int, optional
Number of layers, by default 10
conv_channels : int, optional
Feature size of the convolutional sublayers, by default 64
dilation_factor : int, optional
The factor with which dilation of each convolutional sublayers grows
exponentially if it is greater than 1, else the dilation of each
convolutional sublayers grows linearly, by default 1
nonlinear_activation : str, optional
The activation after each convolutional sublayer, by default "leakyrelu"
nonlinear_activation_params : Dict[str, Any], optional
The parameters passed to the activation's initializer, by default
{"negative_slope": 0.2}
bias : bool, optional
Whether to use bias in convolutional sublayers, by default True
use_weight_norm : bool, optional
Whether to use weight normalization at all convolutional sublayers,
by default True
Args:
in_channels (int, optional): Number of channels of the input audio, by default 1
out_channels (int, optional): Output feature size, by default 1
kernel_size (int, optional): Kernel size of convolutional sublayers, by default 3
layers (int, optional): Number of layers, by default 10
conv_channels (int, optional): Feature size of the convolutional sublayers, by default 64
dilation_factor (int, optional): The factor with which dilation of each convolutional sublayers grows
exponentially if it is greater than 1, else the dilation of each convolutional sublayers grows linearly,
by default 1
nonlinear_activation (str, optional): The activation after each convolutional sublayer, by default "leakyrelu"
nonlinear_activation_params (Dict[str, Any], optional): The parameters passed to the activation's initializer, by default
{"negative_slope": 0.2}
bias (bool, optional): Whether to use bias in convolutional sublayers, by default True
use_weight_norm (bool, optional): Whether to use weight normalization at all convolutional sublayers,
by default True
"""
def __init__(
......@@ -330,15 +288,12 @@ class PWGDiscriminator(nn.Layer):
def forward(self, x):
"""
Parameters
----------
x : Tensor
Shape (N, in_channels, num_samples), the input audio.
Returns
-------
Tensor
Shape (N, out_channels, num_samples), the predicted logits.
Args:
x (Tensor): Shape (N, in_channels, num_samples), the input audio.
Returns:
Tensor: Shape (N, out_channels, num_samples), the predicted logits.
"""
return self.conv_layers(x)
......@@ -362,39 +317,25 @@ class PWGDiscriminator(nn.Layer):
class ResidualPWGDiscriminator(nn.Layer):
"""A wavenet-style discriminator for audio.
Parameters
----------
in_channels : int, optional
Number of channels of the input audio, by default 1
out_channels : int, optional
Output feature size, by default 1
kernel_size : int, optional
Kernel size of residual blocks, by default 3
layers : int, optional
Number of residual blocks, by default 30
stacks : int, optional
Number of groups of residual blocks, within which the dilation
of each residual blocks grows exponentially, by default 3
residual_channels : int, optional
Residual channels of residual blocks, by default 64
gate_channels : int, optional
Gate channels of residual blocks, by default 128
skip_channels : int, optional
Skip channels of residual blocks, by default 64
dropout : float, optional
Dropout probability of residual blocks, by default 0.
bias : bool, optional
Whether to use bias in residual blocks, by default True
use_weight_norm : bool, optional
Whether to use weight normalization in all convolutional layers,
by default True
use_causal_conv : bool, optional
Whether to use causal convolution in residual blocks, by default False
nonlinear_activation : str, optional
Activation after convolutions other than those in residual blocks,
by default "leakyrelu"
nonlinear_activation_params : Dict[str, Any], optional
Parameters to pass to the activation, by default {"negative_slope": 0.2}
Args:
in_channels (int, optional): Number of channels of the input audio, by default 1
out_channels (int, optional): Output feature size, by default 1
kernel_size (int, optional): Kernel size of residual blocks, by default 3
layers (int, optional): Number of residual blocks, by default 30
stacks (int, optional): Number of groups of residual blocks, within which the dilation
of each residual blocks grows exponentially, by default 3
residual_channels (int, optional): Residual channels of residual blocks, by default 64
gate_channels (int, optional): Gate channels of residual blocks, by default 128
skip_channels (int, optional): Skip channels of residual blocks, by default 64
dropout (float, optional): Dropout probability of residual blocks, by default 0.
bias (bool, optional): Whether to use bias in residual blocks, by default True
use_weight_norm (bool, optional): Whether to use weight normalization in all convolutional layers,
by default True
use_causal_conv (bool, optional): Whether to use causal convolution in residual blocks, by default False
nonlinear_activation (str, optional): Activation after convolutions other than those in residual blocks,
by default "leakyrelu"
nonlinear_activation_params (Dict[str, Any], optional): Parameters to pass to the activation,
by default {"negative_slope": 0.2}
"""
def __init__(
......@@ -463,15 +404,11 @@ class ResidualPWGDiscriminator(nn.Layer):
def forward(self, x):
"""
Parameters
----------
x : Tensor
Shape (N, in_channels, num_samples), the input audio.
Returns
-------
Tensor
Shape (N, out_channels, num_samples), the predicted logits.
Args:
x(Tensor): Shape (N, in_channels, num_samples), the input audio.↩
Returns:
Tensor: Shape (N, out_channels, num_samples), the predicted logits.
"""
x = self.first_conv(x)
skip = 0
......
paddlespeech/t2s/models/tacotron2/tacotron2.py
浏览文件 @ 9699c007
......@@ -81,69 +81,39 @@ class Tacotron2(nn.Layer):
# training related
init_type: str="xavier_uniform", ):
"""Initialize Tacotron2 module.
Parameters
----------
idim : int
Dimension of the inputs.
odim : int
Dimension of the outputs.
embed_dim : int
Dimension of the token embedding.
elayers : int
Number of encoder blstm layers.
eunits : int
Number of encoder blstm units.
econv_layers : int
Number of encoder conv layers.
econv_filts : int
Number of encoder conv filter size.
econv_chans : int
Number of encoder conv filter channels.
dlayers : int
Number of decoder lstm layers.
dunits : int
Number of decoder lstm units.
prenet_layers : int
Number of prenet layers.
prenet_units : int
Number of prenet units.
postnet_layers : int
Number of postnet layers.
postnet_filts : int
Number of postnet filter size.
postnet_chans : int
Number of postnet filter channels.
output_activation : str
Name of activation function for outputs.
adim : int
Number of dimension of mlp in attention.
aconv_chans : int
Number of attention conv filter channels.
aconv_filts : int
Number of attention conv filter size.
cumulate_att_w : bool
Whether to cumulate previous attention weight.
use_batch_norm : bool
Whether to use batch normalization.
use_concate : bool
Whether to concat enc outputs w/ dec lstm outputs.
reduction_factor : int
Reduction factor.
spk_num : Optional[int]
Number of speakers. If set to > 1, assume that the
sids will be provided as the input and use sid embedding layer.
lang_num : Optional[int]
Number of languages. If set to > 1, assume that the
lids will be provided as the input and use sid embedding layer.
spk_embed_dim : Optional[int]
Speaker embedding dimension. If set to > 0,
assume that spk_emb will be provided as the input.
spk_embed_integration_type : str
How to integrate speaker embedding.
dropout_rate : float
Dropout rate.
zoneout_rate : float
Zoneout rate.
Args:
idim (int): Dimension of the inputs.
odim (int): Dimension of the outputs.
embed_dim (int): Dimension of the token embedding.
elayers (int): Number of encoder blstm layers.
eunits (int): Number of encoder blstm units.
econv_layers (int): Number of encoder conv layers.
econv_filts (int): Number of encoder conv filter size.
econv_chans (int): Number of encoder conv filter channels.
dlayers (int): Number of decoder lstm layers.
dunits (int): Number of decoder lstm units.
prenet_layers (int): Number of prenet layers.
prenet_units (int): Number of prenet units.
postnet_layers (int): Number of postnet layers.
postnet_filts (int): Number of postnet filter size.
postnet_chans (int): Number of postnet filter channels.
output_activation (str): Name of activation function for outputs.
adim (int): Number of dimension of mlp in attention.
aconv_chans (int): Number of attention conv filter channels.
aconv_filts (int): Number of attention conv filter size.
cumulate_att_w (bool): Whether to cumulate previous attention weight.
use_batch_norm (bool): Whether to use batch normalization.
use_concate (bool): Whether to concat enc outputs w/ dec lstm outputs.
reduction_factor (int): Reduction factor.
spk_num (Optional[int]): Number of speakers. If set to > 1, assume that the
sids will be provided as the input and use sid embedding layer.
lang_num (Optional[int]): Number of languages. If set to > 1, assume that the
lids will be provided as the input and use sid embedding layer.
spk_embed_dim (Optional[int]): Speaker embedding dimension. If set to > 0,
assume that spk_emb will be provided as the input.
spk_embed_integration_type (str): How to integrate speaker embedding.
dropout_rate (float): Dropout rate.
zoneout_rate (float): Zoneout rate.
"""
assert check_argument_types()
super().__init__()
......@@ -258,31 +228,19 @@ class Tacotron2(nn.Layer):
) -> Tuple[paddle.Tensor, Dict[str, paddle.Tensor], paddle.Tensor]:
"""Calculate forward propagation.
Parameters
----------
text : Tensor(int64)
Batch of padded character ids (B, T_text).
text_lengths : Tensor(int64)
Batch of lengths of each input batch (B,).
speech : Tensor
Batch of padded target features (B, T_feats, odim).
speech_lengths : Tensor(int64)
Batch of the lengths of each target (B,).
spk_emb : Optional[Tensor]
Batch of speaker embeddings (B, spk_embed_dim).
spk_id : Optional[Tensor]
Batch of speaker IDs (B, 1).
lang_id : Optional[Tensor]
Batch of language IDs (B, 1).
Returns
----------
Tensor
Loss scalar value.
Dict
Statistics to be monitored.
Tensor
Weight value if not joint training else model outputs.
Args:
text (Tensor(int64)): Batch of padded character ids (B, T_text).
text_lengths (Tensor(int64)): Batch of lengths of each input batch (B,).
speech (Tensor): Batch of padded target features (B, T_feats, odim).
speech_lengths (Tensor(int64)): Batch of the lengths of each target (B,).
spk_emb (Optional[Tensor]): Batch of speaker embeddings (B, spk_embed_dim).
spk_id (Optional[Tensor]): Batch of speaker IDs (B, 1).
lang_id (Optional[Tensor]): Batch of language IDs (B, 1).
Returns:
Tensor: Loss scalar value.
Dict: Statistics to be monitored.
Tensor: Weight value if not joint training else model outputs.
"""
text = text[:, :text_lengths.max()]
......@@ -369,40 +327,26 @@ class Tacotron2(nn.Layer):
use_teacher_forcing: bool=False, ) -> Dict[str, paddle.Tensor]:
"""Generate the sequence of features given the sequences of characters.
Parameters
----------
text Tensor(int64)
Input sequence of characters (T_text,).
speech : Optional[Tensor]
Feature sequence to extract style (N, idim).
spk_emb : ptional[Tensor]
Speaker embedding (spk_embed_dim,).
spk_id : Optional[Tensor]
Speaker ID (1,).
lang_id : Optional[Tensor]
Language ID (1,).
threshold : float
Threshold in inference.
minlenratio : float
Minimum length ratio in inference.
maxlenratio : float
Maximum length ratio in inference.
use_att_constraint : bool
Whether to apply attention constraint.
backward_window : int
Backward window in attention constraint.
forward_window : int
Forward window in attention constraint.
use_teacher_forcing : bool
Whether to use teacher forcing.
Return
----------
Dict[str, Tensor]
Output dict including the following items:
* feat_gen (Tensor): Output sequence of features (T_feats, odim).
* prob (Tensor): Output sequence of stop probabilities (T_feats,).
* att_w (Tensor): Attention weights (T_feats, T).
Args:
text (Tensor(int64)): Input sequence of characters (T_text,).
speech (Optional[Tensor]): Feature sequence to extract style (N, idim).
spk_emb (ptional[Tensor]): Speaker embedding (spk_embed_dim,).
spk_id (Optional[Tensor]): Speaker ID (1,).
lang_id (Optional[Tensor]): Language ID (1,).
threshold (float): Threshold in inference.
minlenratio (float): Minimum length ratio in inference.
maxlenratio (float): Maximum length ratio in inference.
use_att_constraint (bool): Whether to apply attention constraint.
backward_window (int): Backward window in attention constraint.
forward_window (int): Forward window in attention constraint.
use_teacher_forcing (bool): Whether to use teacher forcing.
Returns:
Dict[str, Tensor]
Output dict including the following items:
* feat_gen (Tensor): Output sequence of features (T_feats, odim).
* prob (Tensor): Output sequence of stop probabilities (T_feats,).
* att_w (Tensor): Attention weights (T_feats, T).
"""
x = text
......@@ -458,18 +402,13 @@ class Tacotron2(nn.Layer):
spk_emb: paddle.Tensor) -> paddle.Tensor:
"""Integrate speaker embedding with hidden states.
Parameters
----------
hs : Tensor
Batch of hidden state sequences (B, Tmax, eunits).
spk_emb : Tensor
Batch of speaker embeddings (B, spk_embed_dim).
Returns
----------
Tensor
Batch of integrated hidden state sequences (B, Tmax, eunits) if
integration_type is "add" else (B, Tmax, eunits + spk_embed_dim).
Args:
hs (Tensor): Batch of hidden state sequences (B, Tmax, eunits).
spk_emb (Tensor): Batch of speaker embeddings (B, spk_embed_dim).
Returns:
Tensor: Batch of integrated hidden state sequences (B, Tmax, eunits) if
integration_type is "add" else (B, Tmax, eunits + spk_embed_dim).
"""
if self.spk_embed_integration_type == "add":
......
paddlespeech/t2s/models/transformer_tts/transformer_tts.py
浏览文件 @ 9699c007
......@@ -48,127 +48,67 @@ class TransformerTTS(nn.Layer):
.. _`Neural Speech Synthesis with Transformer Network`:
https://arxiv.org/pdf/1809.08895.pdf
Parameters
----------
idim : int
Dimension of the inputs.
odim : int
Dimension of the outputs.
embed_dim : int, optional
Dimension of character embedding.
eprenet_conv_layers : int, optional
Number of encoder prenet convolution layers.
eprenet_conv_chans : int, optional
Number of encoder prenet convolution channels.
eprenet_conv_filts : int, optional
Filter size of encoder prenet convolution.
dprenet_layers : int, optional
Number of decoder prenet layers.
dprenet_units : int, optional
Number of decoder prenet hidden units.
elayers : int, optional
Number of encoder layers.
eunits : int, optional
Number of encoder hidden units.
adim : int, optional
Number of attention transformation dimensions.
aheads : int, optional
Number of heads for multi head attention.
dlayers : int, optional
Number of decoder layers.
dunits : int, optional
Number of decoder hidden units.
postnet_layers : int, optional
Number of postnet layers.
postnet_chans : int, optional
Number of postnet channels.
postnet_filts : int, optional
Filter size of postnet.
use_scaled_pos_enc : pool, optional
Whether to use trainable scaled positional encoding.
use_batch_norm : bool, optional
Whether to use batch normalization in encoder prenet.
encoder_normalize_before : bool, optional
Whether to perform layer normalization before encoder block.
decoder_normalize_before : bool, optional
Whether to perform layer normalization before decoder block.
encoder_concat_after : bool, optional
Whether to concatenate attention layer's input and output in encoder.
decoder_concat_after : bool, optional
Whether to concatenate attention layer's input and output in decoder.
positionwise_layer_type : str, optional
Position-wise operation type.
positionwise_conv_kernel_size : int, optional
Kernel size in position wise conv 1d.
reduction_factor : int, optional
Reduction factor.
spk_embed_dim : int, optional
Number of speaker embedding dimenstions.
spk_embed_integration_type : str, optional
How to integrate speaker embedding.
use_gst : str, optional
Whether to use global style token.
gst_tokens : int, optional
The number of GST embeddings.
gst_heads : int, optional
The number of heads in GST multihead attention.
gst_conv_layers : int, optional
The number of conv layers in GST.
gst_conv_chans_list : Sequence[int], optional
List of the number of channels of conv layers in GST.
gst_conv_kernel_size : int, optional
Kernal size of conv layers in GST.
gst_conv_stride : int, optional
Stride size of conv layers in GST.
gst_gru_layers : int, optional
The number of GRU layers in GST.
gst_gru_units : int, optional
The number of GRU units in GST.
transformer_lr : float, optional
Initial value of learning rate.
transformer_warmup_steps : int, optional
Optimizer warmup steps.
transformer_enc_dropout_rate : float, optional
Dropout rate in encoder except attention and positional encoding.
transformer_enc_positional_dropout_rate : float, optional
Dropout rate after encoder positional encoding.
transformer_enc_attn_dropout_rate : float, optional
Dropout rate in encoder self-attention module.
transformer_dec_dropout_rate : float, optional
Dropout rate in decoder except attention & positional encoding.
transformer_dec_positional_dropout_rate : float, optional
Dropout rate after decoder positional encoding.
transformer_dec_attn_dropout_rate : float, optional
Dropout rate in deocoder self-attention module.
transformer_enc_dec_attn_dropout_rate : float, optional
Dropout rate in encoder-deocoder attention module.
init_type : str, optional
How to initialize transformer parameters.
init_enc_alpha : float, optional
Initial value of alpha in scaled pos encoding of the encoder.
init_dec_alpha : float, optional
Initial value of alpha in scaled pos encoding of the decoder.
eprenet_dropout_rate : float, optional
Dropout rate in encoder prenet.
dprenet_dropout_rate : float, optional
Dropout rate in decoder prenet.
postnet_dropout_rate : float, optional
Dropout rate in postnet.
use_masking : bool, optional
Whether to apply masking for padded part in loss calculation.
use_weighted_masking : bool, optional
Whether to apply weighted masking in loss calculation.
bce_pos_weight : float, optional
Positive sample weight in bce calculation (only for use_masking=true).
loss_type : str, optional
How to calculate loss.
use_guided_attn_loss : bool, optional
Whether to use guided attention loss.
num_heads_applied_guided_attn : int, optional
Number of heads in each layer to apply guided attention loss.
num_layers_applied_guided_attn : int, optional
Number of layers to apply guided attention loss.
List of module names to apply guided attention loss.
Args:
idim (int): Dimension of the inputs.
odim (int): Dimension of the outputs.
embed_dim (int, optional): Dimension of character embedding.
eprenet_conv_layers (int, optional): Number of encoder prenet convolution layers.
eprenet_conv_chans (int, optional): Number of encoder prenet convolution channels.
eprenet_conv_filts (int, optional): Filter size of encoder prenet convolution.
dprenet_layers (int, optional): Number of decoder prenet layers.
dprenet_units (int, optional): Number of decoder prenet hidden units.
elayers (int, optional): Number of encoder layers.
eunits (int, optional): Number of encoder hidden units.
adim (int, optional): Number of attention transformation dimensions.
aheads (int, optional): Number of heads for multi head attention.
dlayers (int, optional): Number of decoder layers.
dunits (int, optional): Number of decoder hidden units.
postnet_layers (int, optional): Number of postnet layers.
postnet_chans (int, optional): Number of postnet channels.
postnet_filts (int, optional): Filter size of postnet.
use_scaled_pos_enc (pool, optional): Whether to use trainable scaled positional encoding.
use_batch_norm (bool, optional): Whether to use batch normalization in encoder prenet.
encoder_normalize_before (bool, optional): Whether to perform layer normalization before encoder block.
decoder_normalize_before (bool, optional): Whether to perform layer normalization before decoder block.
encoder_concat_after (bool, optional): Whether to concatenate attention layer's input and output in encoder.
decoder_concat_after (bool, optional): Whether to concatenate attention layer's input and output in decoder.
positionwise_layer_type (str, optional): Position-wise operation type.
positionwise_conv_kernel_size (int, optional): Kernel size in position wise conv 1d.
reduction_factor (int, optional): Reduction factor.
spk_embed_dim (int, optional): Number of speaker embedding dimenstions.
spk_embed_integration_type (str, optional): How to integrate speaker embedding.
use_gst (str, optional): Whether to use global style token.
gst_tokens (int, optional): The number of GST embeddings.
gst_heads (int, optional): The number of heads in GST multihead attention.
gst_conv_layers (int, optional): The number of conv layers in GST.
gst_conv_chans_list (Sequence[int], optional): List of the number of channels of conv layers in GST.
gst_conv_kernel_size (int, optional): Kernal size of conv layers in GST.
gst_conv_stride (int, optional): Stride size of conv layers in GST.
gst_gru_layers (int, optional): The number of GRU layers in GST.
gst_gru_units (int, optional): The number of GRU units in GST.
transformer_lr (float, optional): Initial value of learning rate.
transformer_warmup_steps (int, optional): Optimizer warmup steps.
transformer_enc_dropout_rate (float, optional): Dropout rate in encoder except attention and positional encoding.
transformer_enc_positional_dropout_rate (float, optional): Dropout rate after encoder positional encoding.
transformer_enc_attn_dropout_rate （float, optional): Dropout rate in encoder self-attention module.
transformer_dec_dropout_rate (float, optional): Dropout rate in decoder except attention & positional encoding.
transformer_dec_positional_dropout_rate (float, optional): Dropout rate after decoder positional encoding.
transformer_dec_attn_dropout_rate （float, optional): Dropout rate in deocoder self-attention module.
transformer_enc_dec_attn_dropout_rate (float, optional): Dropout rate in encoder-deocoder attention module.
init_type (str, optional): How to initialize transformer parameters.
init_enc_alpha （float, optional）: Initial value of alpha in scaled pos encoding of the encoder.
init_dec_alpha (float, optional): Initial value of alpha in scaled pos encoding of the decoder.
eprenet_dropout_rate (float, optional): Dropout rate in encoder prenet.
dprenet_dropout_rate (float, optional): Dropout rate in decoder prenet.
postnet_dropout_rate (float, optional): Dropout rate in postnet.
use_masking (bool, optional): Whether to apply masking for padded part in loss calculation.
use_weighted_masking (bool, optional): Whether to apply weighted masking in loss calculation.
bce_pos_weight (float, optional): Positive sample weight in bce calculation (only for use_masking=true).
loss_type (str, optional): How to calculate loss.
use_guided_attn_loss (bool, optional): Whether to use guided attention loss.
num_heads_applied_guided_attn (int, optional): Number of heads in each layer to apply guided attention loss.
num_layers_applied_guided_attn (int, optional): Number of layers to apply guided attention loss.
List of module names to apply guided attention loss.
"""
def __init__(
......@@ -398,25 +338,16 @@ class TransformerTTS(nn.Layer):
) -> Tuple[paddle.Tensor, Dict[str, paddle.Tensor], paddle.Tensor]:
"""Calculate forward propagation.
Parameters
----------
text : Tensor(int64)
Batch of padded character ids (B, Tmax).
text_lengths : Tensor(int64)
Batch of lengths of each input batch (B,).
speech : Tensor
Batch of padded target features (B, Lmax, odim).
speech_lengths : Tensor(int64)
Batch of the lengths of each target (B,).
spk_emb : Tensor, optional
Batch of speaker embeddings (B, spk_embed_dim).
Returns
----------
Tensor
Loss scalar value.
Dict
Statistics to be monitored.
Args:
text(Tensor(int64)): Batch of padded character ids (B, Tmax).
text_lengths(Tensor(int64)): Batch of lengths of each input batch (B,).
speech(Tensor): Batch of padded target features (B, Lmax, odim).
speech_lengths(Tensor(int64)): Batch of the lengths of each target (B,).
spk_emb(Tensor, optional): Batch of speaker embeddings (B, spk_embed_dim).
Returns:
Tensor: Loss scalar value.
Dict: Statistics to be monitored.
"""
# input of embedding must be int64
......@@ -525,31 +456,19 @@ class TransformerTTS(nn.Layer):
) -> Tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor]:
"""Generate the sequence of features given the sequences of characters.
Parameters
----------
text : Tensor(int64)
Input sequence of characters (T,).
speech : Tensor, optional
Feature sequence to extract style (N, idim).
spk_emb : Tensor, optional
Speaker embedding vector (spk_embed_dim,).
threshold : float, optional
Threshold in inference.
minlenratio : float, optional
Minimum length ratio in inference.
maxlenratio : float, optional
Maximum length ratio in inference.
use_teacher_forcing : bool, optional
Whether to use teacher forcing.
Returns
----------
Tensor
Output sequence of features (L, odim).
Tensor
Output sequence of stop probabilities (L,).
Tensor
Encoder-decoder (source) attention weights (#layers, #heads, L, T).
Args:
text(Tensor(int64)): Input sequence of characters (T,).
speech(Tensor, optional): Feature sequence to extract style (N, idim).
spk_emb(Tensor, optional): Speaker embedding vector (spk_embed_dim,).
threshold(float, optional): Threshold in inference.
minlenratio(float, optional): Minimum length ratio in inference.
maxlenratio(float, optional): Maximum length ratio in inference.
use_teacher_forcing(bool, optional): Whether to use teacher forcing.
Returns:
Tensor: Output sequence of features (L, odim).
Tensor: Output sequence of stop probabilities (L,).
Tensor: Encoder-decoder (source) attention weights (#layers, #heads, L, T).
"""
# input of embedding must be int64
......@@ -671,23 +590,17 @@ class TransformerTTS(nn.Layer):
def _source_mask(self, ilens: paddle.Tensor) -> paddle.Tensor:
"""Make masks for self-attention.
Parameters
----------
ilens : Tensor
Batch of lengths (B,).
Args:
ilens(Tensor): Batch of lengths (B,).
Returns
-------
Tensor
Mask tensor for self-attention.
dtype=paddle.bool
Returns:
Tensor: Mask tensor for self-attention. dtype=paddle.bool
Examples
-------
>>> ilens = [5, 3]
>>> self._source_mask(ilens)
tensor([[[1, 1, 1, 1, 1],
[1, 1, 1, 0, 0]]]) bool
Examples:
>>> ilens = [5, 3]
>>> self._source_mask(ilens)
tensor([[[1, 1, 1, 1, 1],
[1, 1, 1, 0, 0]]]) bool
"""
x_masks = make_non_pad_mask(ilens)
......@@ -696,30 +609,25 @@ class TransformerTTS(nn.Layer):
def _target_mask(self, olens: paddle.Tensor) -> paddle.Tensor:
"""Make masks for masked self-attention.
Parameters
----------
olens : LongTensor
Batch of lengths (B,).
Returns
----------
Tensor
Mask tensor for masked self-attention.
Examples
----------
>>> olens = [5, 3]
>>> self._target_mask(olens)
tensor([[[1, 0, 0, 0, 0],
[1, 1, 0, 0, 0],
[1, 1, 1, 0, 0],
[1, 1, 1, 1, 0],
[1, 1, 1, 1, 1]],
[[1, 0, 0, 0, 0],
[1, 1, 0, 0, 0],
[1, 1, 1, 0, 0],
[1, 1, 1, 0, 0],
[1, 1, 1, 0, 0]]], dtype=paddle.uint8)
Args:
olens (Tensor(int64)): Batch of lengths (B,).
Returns:
Tensor: Mask tensor for masked self-attention.
Examples:
>>> olens = [5, 3]
>>> self._target_mask(olens)
tensor([[[1, 0, 0, 0, 0],
[1, 1, 0, 0, 0],
[1, 1, 1, 0, 0],
[1, 1, 1, 1, 0],
[1, 1, 1, 1, 1]],
[[1, 0, 0, 0, 0],
[1, 1, 0, 0, 0],
[1, 1, 1, 0, 0],
[1, 1, 1, 0, 0],
[1, 1, 1, 0, 0]]], dtype=paddle.uint8)
"""
y_masks = make_non_pad_mask(olens)
......@@ -731,17 +639,12 @@ class TransformerTTS(nn.Layer):
spk_emb: paddle.Tensor) -> paddle.Tensor:
"""Integrate speaker embedding with hidden states.
Parameters
----------
hs : Tensor
Batch of hidden state sequences (B, Tmax, adim).
spk_emb : Tensor
Batch of speaker embeddings (B, spk_embed_dim).
Returns
----------
Tensor
Batch of integrated hidden state sequences (B, Tmax, adim).
Args:
hs(Tensor): Batch of hidden state sequences (B, Tmax, adim).
spk_emb(Tensor): Batch of speaker embeddings (B, spk_embed_dim).
Returns:
Tensor: Batch of integrated hidden state sequences (B, Tmax, adim).
"""
if self.spk_embed_integration_type == "add":
......
paddlespeech/t2s/models/waveflow.py
浏览文件 @ 9699c007
此差异已折叠。
paddlespeech/t2s/models/wavernn/wavernn.py
浏览文件 @ 9699c007
......@@ -67,14 +67,10 @@ class MelResNet(nn.Layer):
def forward(self, x):
'''
Parameters
----------
x : Tensor
Input tensor (B, in_dims, T).
Returns
----------
Tensor
Output tensor (B, res_out_dims, T).
Args:
x (Tensor): Input tensor (B, in_dims, T).
Returns:
Tensor: Output tensor (B, res_out_dims, T).
'''
x = self.conv_in(x)
......@@ -121,16 +117,11 @@ class UpsampleNetwork(nn.Layer):
def forward(self, m):
'''
Parameters
----------
c : Tensor
Input tensor (B, C_aux, T).
Returns
----------
Tensor
Output tensor (B, (T - 2 * pad) * prob(upsample_scales), C_aux).
Tensor
Output tensor (B, (T - 2 * pad) * prob(upsample_scales), res_out_dims).
Args:
c (Tensor): Input tensor (B, C_aux, T).
Returns:
Tensor: Output tensor (B, (T - 2 * pad) * prob(upsample_scales), C_aux).
Tensor: Output tensor (B, (T - 2 * pad) * prob(upsample_scales), res_out_dims).
'''
# aux: [B, C_aux, T]
# -> [B, res_out_dims, T - 2 * aux_context_window]
......@@ -172,32 +163,20 @@ class WaveRNN(nn.Layer):
mode='RAW',
init_type: str="xavier_uniform", ):
'''
Parameters
----------
rnn_dims : int, optional
Hidden dims of RNN Layers.
fc_dims : int, optional
Dims of FC Layers.
bits : int, optional
bit depth of signal.
aux_context_window : int, optional
The context window size of the first convolution applied to the
auxiliary input, by default 2
upsample_scales : List[int], optional
Upsample scales of the upsample network.
aux_channels : int, optional
Auxiliary channel of the residual blocks.
compute_dims : int, optional
Dims of Conv1D in MelResNet.
res_out_dims : int, optional
Dims of output in MelResNet.
res_blocks : int, optional
Number of residual blocks.
mode : str, optional
Output mode of the WaveRNN vocoder. `MOL` for Mixture of Logistic Distribution,
and `RAW` for quantized bits as the model's output.
init_type : str
How to initialize parameters.
Args:
rnn_dims (int, optional): Hidden dims of RNN Layers.
fc_dims (int, optional): Dims of FC Layers.
bits (int, optional): bit depth of signal.
aux_context_window (int, optional): The context window size of the first convolution applied to the
auxiliary input, by default 2
upsample_scales (List[int], optional): Upsample scales of the upsample network.
aux_channels (int, optional): Auxiliary channel of the residual blocks.
compute_dims (int, optional): Dims of Conv1D in MelResNet.
res_out_dims (int, optional): Dims of output in MelResNet.
res_blocks (int, optional): Number of residual blocks.
mode (str, optional): Output mode of the WaveRNN vocoder.
`MOL` for Mixture of Logistic Distribution, and `RAW` for quantized bits as the model's output.
init_type (str): How to initialize parameters.
'''
super().__init__()
self.mode = mode
......@@ -245,18 +224,13 @@ class WaveRNN(nn.Layer):
def forward(self, x, c):
'''
Parameters
----------
x : Tensor
wav sequence, [B, T]
c : Tensor
mel spectrogram [B, C_aux, T']
T = (T' - 2 * aux_context_window ) * hop_length
Returns
----------
Tensor
[B, T, n_classes]
Args:
x (Tensor): wav sequence, [B, T]
c (Tensor): mel spectrogram [B, C_aux, T']
T = (T' - 2 * aux_context_window ) * hop_length
Returns:
Tensor: [B, T, n_classes]
'''
# Although we `_flatten_parameters()` on init, when using DataParallel
# the model gets replicated, making it no longer guaranteed that the
......@@ -304,22 +278,14 @@ class WaveRNN(nn.Layer):
mu_law: bool=True,
gen_display: bool=False):
"""
Parameters
----------
c : Tensor
input mels, (T', C_aux)
batched : bool
generate in batch or not
target : int
target number of samples to be generated in each batch entry
overlap : int
number of samples for crossfading between batches
mu_law : bool
use mu law or not
Returns
----------
wav sequence
Output (T' * prod(upsample_scales), out_channels, C_out).
Args:
c(Tensor): input mels, (T', C_aux)
batched(bool): generate in batch or not
target(int): target number of samples to be generated in each batch entry
overlap(int): number of samples for crossfading between batches
mu_law(bool)
Returns:
wav sequence: Output (T' * prod(upsample_scales), out_channels, C_out).
"""
self.eval()
......@@ -434,16 +400,13 @@ class WaveRNN(nn.Layer):
def pad_tensor(self, x, pad, side='both'):
'''
Parameters
----------
x : Tensor
mel, [1, n_frames, 80]
pad : int
side : str
'both', 'before' or 'after'
Returns
----------
Tensor
Args:
x(Tensor): mel, [1, n_frames, 80]
pad(int):
side(str, optional): (Default value = 'both')
Returns:
Tensor
'''
b, t, _ = paddle.shape(x)
# for dygraph to static graph
......@@ -461,38 +424,29 @@ class WaveRNN(nn.Layer):
Fold the tensor with overlap for quick batched inference.
Overlap will be used for crossfading in xfade_and_unfold()
Parameters
----------
x : Tensor
Upsampled conditioning features. mels or aux
shape=(1, T, features)
mels: [1, T, 80]
aux: [1, T, 128]
target : int
Target timesteps for each index of batch
overlap : int
Timesteps for both xfade and rnn warmup
overlap = hop_length * 2
Returns
----------
Tensor
shape=(num_folds, target + 2 * overlap, features)
num_flods = (time_seq - overlap) // (target + overlap)
mel: [num_folds, target + 2 * overlap, 80]
aux: [num_folds, target + 2 * overlap, 128]
Details
----------
x = [[h1, h2, ... hn]]
Where each h is a vector of conditioning features
Eg: target=2, overlap=1 with x.size(1)=10
folded = [[h1, h2, h3, h4],
[h4, h5, h6, h7],
[h7, h8, h9, h10]]
Args:
x(Tensor): Upsampled conditioning features. mels or aux
shape=(1, T, features)
mels: [1, T, 80]
aux: [1, T, 128]
target(int): Target timesteps for each index of batch
overlap(int): Timesteps for both xfade and rnn warmup
Returns:
Tensor:
shape=(num_folds, target + 2 * overlap, features)
num_flods = (time_seq - overlap) // (target + overlap)
mel: [num_folds, target + 2 * overlap, 80]
aux: [num_folds, target + 2 * overlap, 128]
Details:
x = [[h1, h2, ... hn]]
Where each h is a vector of conditioning features
Eg: target=2, overlap=1 with x.size(1)=10
folded = [[h1, h2, h3, h4],
[h4, h5, h6, h7],
[h7, h8, h9, h10]]
'''
_, total_len, features = paddle.shape(x)
......@@ -520,37 +474,33 @@ class WaveRNN(nn.Layer):
def xfade_and_unfold(self, y, target: int=12000, overlap: int=600):
''' Applies a crossfade and unfolds into a 1d array.
Parameters
----------
y : Tensor
Batched sequences of audio samples
shape=(num_folds, target + 2 * overlap)
dtype=paddle.float32
overlap : int
Timesteps for both xfade and rnn warmup
Returns
----------
Tensor
audio samples in a 1d array
shape=(total_len)
dtype=paddle.float32
Details
----------
y = [[seq1],
[seq2],
[seq3]]
Apply a gain envelope at both ends of the sequences
y = [[seq1_in, seq1_target, seq1_out],
[seq2_in, seq2_target, seq2_out],
[seq3_in, seq3_target, seq3_out]]
Stagger and add up the groups of samples:
[seq1_in, seq1_target, (seq1_out + seq2_in), seq2_target, ...]
Args:
y (Tensor):
Batched sequences of audio samples
shape=(num_folds, target + 2 * overlap)
dtype=paddle.float32
overlap (int): Timesteps for both xfade and rnn warmup
Returns:
Tensor
audio samples in a 1d array
shape=(total_len)
dtype=paddle.float32
Details:
y = [[seq1],
[seq2],
[seq3]]
Apply a gain envelope at both ends of the sequences
y = [[seq1_in, seq1_target, seq1_out],
[seq2_in, seq2_target, seq2_out],
[seq3_in, seq3_target, seq3_out]]
Stagger and add up the groups of samples:
[seq1_in, seq1_target, (seq1_out + seq2_in), seq2_target, ...]
'''
# num_folds = (total_len - overlap) // (target + overlap)
......
paddlespeech/t2s/modules/causal_conv.py
浏览文件 @ 9699c007
......@@ -41,14 +41,10 @@ class CausalConv1D(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input tensor (B, in_channels, T).
Returns
----------
Tensor
Output tensor (B, out_channels, T).
Args:
x (Tensor): Input tensor (B, in_channels, T).
Returns:
Tensor: Output tensor (B, out_channels, T).
"""
return self.conv(self.pad(x))[:, :, :x.shape[2]]
......@@ -70,13 +66,9 @@ class CausalConv1DTranspose(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input tensor (B, in_channels, T_in).
Returns
----------
Tensor
Output tensor (B, out_channels, T_out).
Args:
x (Tensor): Input tensor (B, in_channels, T_in).
Returns:
Tensor: Output tensor (B, out_channels, T_out).
"""
return self.deconv(x)[:, :, :-self.stride]
paddlespeech/t2s/modules/conformer/convolution.py
浏览文件 @ 9699c007
......@@ -18,12 +18,10 @@ from paddle import nn
class ConvolutionModule(nn.Layer):
"""ConvolutionModule in Conformer model.
Parameters
----------
channels : int
The number of channels of conv layers.
kernel_size : int
Kernerl size of conv layers.
Args:
channels (int): The number of channels of conv layers.
kernel_size (int): Kernerl size of conv layers.
"""
def __init__(self, channels, kernel_size, activation=nn.ReLU(), bias=True):
......@@ -59,14 +57,11 @@ class ConvolutionModule(nn.Layer):
def forward(self, x):
"""Compute convolution module.
Parameters
----------
x : paddle.Tensor
Input tensor (#batch, time, channels).
Returns
----------
paddle.Tensor
Output tensor (#batch, time, channels).
Args:
x (Tensor): Input tensor (#batch, time, channels).
Returns:
Tensor: Output tensor (#batch, time, channels).
"""
# exchange the temporal dimension and the feature dimension
x = x.transpose([0, 2, 1])
......
paddlespeech/t2s/modules/conformer/encoder_layer.py
浏览文件 @ 9699c007
......@@ -21,38 +21,29 @@ from paddlespeech.t2s.modules.layer_norm import LayerNorm
class EncoderLayer(nn.Layer):
"""Encoder layer module.
Parameters
----------
size : int
Input dimension.
self_attn : nn.Layer
Self-attention module instance.
`MultiHeadedAttention` or `RelPositionMultiHeadedAttention` instance
can be used as the argument.
feed_forward : nn.Layer
Feed-forward module instance.
`PositionwiseFeedForward`, `MultiLayeredConv1d`, or `Conv1dLinear` instance
can be used as the argument.
feed_forward_macaron : nn.Layer
Additional feed-forward module instance.
`PositionwiseFeedForward`, `MultiLayeredConv1d`, or `Conv1dLinear` instance
can be used as the argument.
conv_module : nn.Layer
Convolution module instance.
`ConvlutionModule` instance can be used as the argument.
dropout_rate : float
Dropout rate.
normalize_before : bool
Whether to use layer_norm before the first block.
concat_after : bool
Whether to concat attention layer's input and output.
if True, additional linear will be applied.
i.e. x -> x + linear(concat(x, att(x)))
if False, no additional linear will be applied. i.e. x -> x + att(x)
stochastic_depth_rate : float
Proability to skip this layer.
During training, the layer may skip residual computation and return input
as-is with given probability.
Args:
size (int): Input dimension.
self_attn (nn.Layer): Self-attention module instance.
`MultiHeadedAttention` or `RelPositionMultiHeadedAttention` instance
can be used as the argument.
feed_forward (nn.Layer): Feed-forward module instance.
`PositionwiseFeedForward`, `MultiLayeredConv1d`, or `Conv1dLinear` instance
can be used as the argument.
feed_forward_macaron (nn.Layer): Additional feed-forward module instance.
`PositionwiseFeedForward`, `MultiLayeredConv1d`, or `Conv1dLinear` instance
can be used as the argument.
conv_module (nn.Layer): Convolution module instance.
`ConvlutionModule` instance can be used as the argument.
dropout_rate (float): Dropout rate.
normalize_before (bool): Whether to use layer_norm before the first block.
concat_after (bool): Whether to concat attention layer's input and output.
if True, additional linear will be applied.
i.e. x -> x + linear(concat(x, att(x)))
if False, no additional linear will be applied. i.e. x -> x + att(x)
stochastic_depth_rate (float): Proability to skip this layer.
During training, the layer may skip residual computation and return input
as-is with given probability.
"""
def __init__(
......@@ -93,22 +84,17 @@ class EncoderLayer(nn.Layer):
def forward(self, x_input, mask, cache=None):
"""Compute encoded features.
Parameters
----------
x_input : Union[Tuple, paddle.Tensor]
Input tensor w/ or w/o pos emb.
- w/ pos emb: Tuple of tensors [(#batch, time, size), (1, time, size)].
- w/o pos emb: Tensor (#batch, time, size).
mask : paddle.Tensor
Mask tensor for the input (#batch, time).
cache paddle.Tensor
Cache tensor of the input (#batch, time - 1, size).
Returns
----------
paddle.Tensor
Output tensor (#batch, time, size).
paddle.Tensor
Mask tensor (#batch, time).
Args:
x_input(Union[Tuple, Tensor]): Input tensor w/ or w/o pos emb.
- w/ pos emb: Tuple of tensors [(#batch, time, size), (1, time, size)].
- w/o pos emb: Tensor (#batch, time, size).
mask(Tensor): Mask tensor for the input (#batch, time).
cache (Tensor):
Returns:
Tensor: Output tensor (#batch, time, size).
Tensor: Mask tensor (#batch, time).
"""
if isinstance(x_input, tuple):
x, pos_emb = x_input[0], x_input[1]
......
paddlespeech/t2s/modules/conv.py
浏览文件 @ 9699c007
......@@ -40,36 +40,29 @@ class Conv1dCell(nn.Conv1D):
2. padding must be a causal padding (recpetive_field - 1, 0).
Thus, these arguments are removed from the ``__init__`` method of this
class.
Parameters
----------
in_channels: int
The feature size of the input.
out_channels: int
The feature size of the output.
kernel_size: int or Tuple[int]
The size of the kernel.
dilation: int or Tuple[int]
The dilation of the convolution, by default 1
weight_attr: ParamAttr, Initializer, str or bool, optional
The parameter attribute of the convolution kernel, by default None.
bias_attr: ParamAttr, Initializer, str or bool, optional
The parameter attribute of the bias. If ``False``, this layer does not
have a bias, by default None.
Examples
--------
>>> cell = Conv1dCell(3, 4, kernel_size=5)
>>> inputs = [paddle.randn([4, 3]) for _ in range(16)]
>>> outputs = []
>>> cell.eval()
>>> cell.start_sequence()
>>> for xt in inputs:
>>> outputs.append(cell.add_input(xt))
>>> len(outputs))
16
>>> outputs[0].shape
[4, 4]
Args:
in_channels (int): The feature size of the input.
out_channels (int): The feature size of the output.
kernel_size (int or Tuple[int]): The size of the kernel.
dilation (int or Tuple[int]): The dilation of the convolution, by default 1
weight_attr (ParamAttr, Initializer, str or bool, optional) : The parameter attribute of the convolution kernel,
by default None.
bias_attr (ParamAttr, Initializer, str or bool, optional):The parameter attribute of the bias.
If ``False``, this layer does not have a bias, by default None.
Examples:
>>> cell = Conv1dCell(3, 4, kernel_size=5)
>>> inputs = [paddle.randn([4, 3]) for _ in range(16)]
>>> outputs = []
>>> cell.eval()
>>> cell.start_sequence()
>>> for xt in inputs:
>>> outputs.append(cell.add_input(xt))
>>> len(outputs))
16
>>> outputs[0].shape
[4, 4]
"""
def __init__(self,
......@@ -103,15 +96,13 @@ class Conv1dCell(nn.Conv1D):
def start_sequence(self):
"""Prepare the layer for a series of incremental forward.
Warnings
---------
This method should be called before a sequence of calls to
``add_input``.
Warnings:
This method should be called before a sequence of calls to
``add_input``.
Raises
------
Exception
If this method is called when the layer is in training mode.
Raises:
Exception
If this method is called when the layer is in training mode.
"""
if self.training:
raise Exception("only use start_sequence in evaluation")
......@@ -130,10 +121,9 @@ class Conv1dCell(nn.Conv1D):
def initialize_buffer(self, x_t):
"""Initialize the buffer for the step input.
Parameters
----------
x_t : Tensor [shape=(batch_size, in_channels)]
The step input.
Args:
x_t (Tensor): The step input. shape=(batch_size, in_channels)
"""
batch_size, _ = x_t.shape
self._buffer = paddle.zeros(
......@@ -143,26 +133,22 @@ class Conv1dCell(nn.Conv1D):
def update_buffer(self, x_t):
"""Shift the buffer by one step.
Parameters
----------
x_t : Tensor [shape=(batch_size, in_channels)]
The step input.
Args:
x_t (Tensor): The step input. shape=(batch_size, in_channels)
"""
self._buffer = paddle.concat(
[self._buffer[:, :, 1:], paddle.unsqueeze(x_t, -1)], -1)
def add_input(self, x_t):
"""Add step input and compute step output.
Parameters
-----------
x_t : Tensor [shape=(batch_size, in_channels)]
The step input.
Returns
-------
y_t :Tensor [shape=(batch_size, out_channels)]
The step output.
Args:
x_t (Tensor): The step input. shape=(batch_size, in_channels)
Returns:
y_t (Tensor): The step output. shape=(batch_size, out_channels)
"""
batch_size = x_t.shape[0]
if self.receptive_field > 1:
......@@ -186,33 +172,26 @@ class Conv1dCell(nn.Conv1D):
class Conv1dBatchNorm(nn.Layer):
"""A Conv1D Layer followed by a BatchNorm1D.
Parameters
----------
in_channels : int
The feature size of the input.
out_channels : int
The feature size of the output.
kernel_size : int
The size of the convolution kernel.
stride : int, optional
The stride of the convolution, by default 1.
padding : int, str or Tuple[int], optional
The padding of the convolution.
If int, a symmetrical padding is applied before convolution;
If str, it should be "same" or "valid";
If Tuple[int], its length should be 2, meaning
``(pad_before, pad_after)``, by default 0.
weight_attr : ParamAttr, Initializer, str or bool, optional
The parameter attribute of the convolution kernel, by default None.
bias_attr : ParamAttr, Initializer, str or bool, optional
The parameter attribute of the bias of the convolution, by default
None.
data_format : str ["NCL" or "NLC"], optional
The data layout of the input, by default "NCL"
momentum : float, optional
The momentum of the BatchNorm1D layer, by default 0.9
epsilon : [type], optional
The epsilon of the BatchNorm1D layer, by default 1e-05
Args:
in_channels (int): The feature size of the input.
out_channels (int): The feature size of the output.
kernel_size (int): The size of the convolution kernel.
stride (int, optional): The stride of the convolution, by default 1.
padding (int, str or Tuple[int], optional):
The padding of the convolution.
If int, a symmetrical padding is applied before convolution;
If str, it should be "same" or "valid";
If Tuple[int], its length should be 2, meaning
``(pad_before, pad_after)``, by default 0.
weight_attr (ParamAttr, Initializer, str or bool, optional):
The parameter attribute of the convolution kernel,
by default None.
bias_attr (ParamAttr, Initializer, str or bool, optional):
The parameter attribute of the bias of the convolution,
by defaultNone.
data_format (str ["NCL" or "NLC"], optional): The data layout of the input, by default "NCL"
momentum (float, optional): The momentum of the BatchNorm1D layer, by default 0.9
epsilon (float, optional): The epsilon of the BatchNorm1D layer, by default 1e-05
"""
def __init__(self,
......@@ -244,16 +223,15 @@ class Conv1dBatchNorm(nn.Layer):
def forward(self, x):
"""Forward pass of the Conv1dBatchNorm layer.
Parameters
----------
x : Tensor [shape=(B, C_in, T_in) or (B, T_in, C_in)]
The input tensor. Its data layout depends on ``data_format``.
Returns
-------
Tensor [shape=(B, C_out, T_out) or (B, T_out, C_out)]
The output tensor.
Args:
x (Tensor): The input tensor. Its data layout depends on ``data_format``.
shape=(B, C_in, T_in) or (B, T_in, C_in)
Returns:
Tensor: The output tensor.
shape=(B, C_out, T_out) or (B, T_out, C_out)
"""
x = self.conv(x)
x = self.bn(x)
......
paddlespeech/t2s/modules/geometry.py
浏览文件 @ 9699c007
......@@ -17,24 +17,18 @@ import paddle
def shuffle_dim(x, axis, perm=None):
"""Permute input tensor along aixs given the permutation or randomly.
Args:
x (Tensor): The input tensor.
axis (int): The axis to shuffle.
perm (List[int], ndarray, optional):
The order to reorder the tensor along the ``axis``-th dimension.
It is a permutation of ``[0, d)``, where d is the size of the
``axis``-th dimension of the input tensor. If not provided,
a random permutation is used. Defaults to None.
Parameters
----------
x : Tensor
The input tensor.
axis : int
The axis to shuffle.
perm : List[int], ndarray, optional
The order to reorder the tensor along the ``axis``-th dimension.
It is a permutation of ``[0, d)``, where d is the size of the
``axis``-th dimension of the input tensor. If not provided,
a random permutation is used. Defaults to None.
Returns
---------
Tensor
The shuffled tensor, which has the same shape as x does.
Returns:
Tensor: The shuffled tensor, which has the same shape as x does.
"""
size = x.shape[axis]
if perm is not None and len(perm) != size:
......
paddlespeech/t2s/modules/layer_norm.py
浏览文件 @ 9699c007
......@@ -18,13 +18,9 @@ from paddle import nn
class LayerNorm(nn.LayerNorm):
"""Layer normalization module.
Parameters
----------
nout : int
Output dim size.
dim : int
Dimension to be normalized.
Args:
nout (int): Output dim size.
dim (int): Dimension to be normalized.
"""
def __init__(self, nout, dim=-1):
......@@ -35,15 +31,11 @@ class LayerNorm(nn.LayerNorm):
def forward(self, x):
"""Apply layer normalization.
Parameters
----------
x : paddle.Tensor
Input tensor.
Args:
x (Tensor):Input tensor.
Returns
----------
paddle.Tensor
Normalized tensor.
Returns:
Tensor: Normalized tensor.
"""
if self.dim == -1:
......
paddlespeech/t2s/modules/losses.py
浏览文件 @ 9699c007
此差异已折叠。
paddlespeech/t2s/modules/nets_utils.py
浏览文件 @ 9699c007
......@@ -20,27 +20,21 @@ from typeguard import check_argument_types
def pad_list(xs, pad_value):
"""Perform padding for the list of tensors.
Parameters
----------
xs : List[Tensor]
List of Tensors [(T_1, `*`), (T_2, `*`), ..., (T_B, `*`)].
pad_value : float)
Value for padding.
Returns
----------
Tensor
Padded tensor (B, Tmax, `*`).
Examples
----------
>>> x = [paddle.ones([4]), paddle.ones([2]), paddle.ones([1])]
>>> x
[tensor([1., 1., 1., 1.]), tensor([1., 1.]), tensor([1.])]
>>> pad_list(x, 0)
tensor([[1., 1., 1., 1.],
[1., 1., 0., 0.],
[1., 0., 0., 0.]])
Args:
xs (List[Tensor]): List of Tensors [(T_1, `*`), (T_2, `*`), ..., (T_B, `*`)].
pad_value (float): Value for padding.
Returns:
Tensor: Padded tensor (B, Tmax, `*`).
Examples:
>>> x = [paddle.ones([4]), paddle.ones([2]), paddle.ones([1])]
>>> x
[tensor([1., 1., 1., 1.]), tensor([1., 1.]), tensor([1.])]
>>> pad_list(x, 0)
tensor([[1., 1., 1., 1.],
[1., 1., 0., 0.],
[1., 0., 0., 0.]])
"""
n_batch = len(xs)
max_len = max(x.shape[0] for x in xs)
......@@ -55,25 +49,20 @@ def pad_list(xs, pad_value):
def make_pad_mask(lengths, length_dim=-1):
"""Make mask tensor containing indices of padded part.
Parameters
----------
lengths : LongTensor
Batch of lengths (B,).
Returns
----------
Tensor(bool)
Mask tensor containing indices of padded part bool.
Examples
----------
With only lengths.
>>> lengths = [5, 3, 2]
>>> make_non_pad_mask(lengths)
masks = [[0, 0, 0, 0 ,0],
[0, 0, 0, 1, 1],
[0, 0, 1, 1, 1]]
Args:
lengths (Tensor(int64)): Batch of lengths (B,).
Returns:
Tensor(bool): Mask tensor containing indices of padded part bool.
Examples:
With only lengths.
>>> lengths = [5, 3, 2]
>>> make_non_pad_mask(lengths)
masks = [[0, 0, 0, 0 ,0],
[0, 0, 0, 1, 1],
[0, 0, 1, 1, 1]]
"""
if length_dim == 0:
raise ValueError("length_dim cannot be 0: {}".format(length_dim))
......@@ -91,31 +80,24 @@ def make_pad_mask(lengths, length_dim=-1):
def make_non_pad_mask(lengths, length_dim=-1):
"""Make mask tensor containing indices of non-padded part.
Parameters
----------
lengths : LongTensor or List
Batch of lengths (B,).
xs : Tensor, optional
The reference tensor.
If set, masks will be the same shape as this tensor.
length_dim : int, optional
Dimension indicator of the above tensor.
See the example.
Returns
----------
Tensor(bool)
mask tensor containing indices of padded part bool.
Examples
----------
With only lengths.
>>> lengths = [5, 3, 2]
>>> make_non_pad_mask(lengths)
masks = [[1, 1, 1, 1 ,1],
[1, 1, 1, 0, 0],
[1, 1, 0, 0, 0]]
Args:
lengths (Tensor(int64) or List): Batch of lengths (B,).
xs (Tensor, optional): The reference tensor.
If set, masks will be the same shape as this tensor.
length_dim (int, optional): Dimension indicator of the above tensor.
See the example.
Returns:
Tensor(bool): mask tensor containing indices of padded part bool.
Examples:
With only lengths.
>>> lengths = [5, 3, 2]
>>> make_non_pad_mask(lengths)
masks = [[1, 1, 1, 1 ,1],
[1, 1, 1, 0, 0],
[1, 1, 0, 0, 0]]
"""
return paddle.logical_not(make_pad_mask(lengths, length_dim))
......@@ -127,12 +109,9 @@ def initialize(model: nn.Layer, init: str):
Custom initialization routines can be implemented into submodules
Parameters
----------
model : nn.Layer
Target.
init : str
Method of initialization.
Args:
model (nn.Layer): Target.
init (str): Method of initialization.
"""
assert check_argument_types()
......
paddlespeech/t2s/modules/pqmf.py
浏览文件 @ 9699c007
......@@ -24,20 +24,16 @@ def design_prototype_filter(taps=62, cutoff_ratio=0.142, beta=9.0):
"""Design prototype filter for PQMF.
This method is based on `A Kaiser window approach for the design of prototype
filters of cosine modulated filterbanks`_.
Parameters
----------
taps : int
The number of filter taps.
cutoff_ratio : float
Cut-off frequency ratio.
beta : float
Beta coefficient for kaiser window.
Returns
----------
ndarray
Impluse response of prototype filter (taps + 1,).
.. _`A Kaiser window approach for the design of prototype filters of cosine modulated filterbanks`:
https://ieeexplore.ieee.org/abstract/document/681427
Args:
taps (int): The number of filter taps.
cutoff_ratio (float): Cut-off frequency ratio.
beta (float): Beta coefficient for kaiser window.
Returns:
ndarray:
Impluse response of prototype filter (taps + 1,).
.. _`A Kaiser window approach for the design of prototype filters of cosine modulated filterbanks`:
https://ieeexplore.ieee.org/abstract/document/681427
"""
# check the arguments are valid
assert taps % 2 == 0, "The number of taps mush be even number."
......@@ -68,16 +64,12 @@ class PQMF(nn.Layer):
"""Initilize PQMF module.
The cutoff_ratio and beta parameters are optimized for #subbands = 4.
See dicussion in https://github.com/kan-bayashi/ParallelWaveGAN/issues/195.
Parameters
----------
subbands : int
The number of subbands.
taps : int
The number of filter taps.
cutoff_ratio : float
Cut-off frequency ratio.
beta : float
Beta coefficient for kaiser window.
Args:
subbands (int): The number of subbands.
taps (int): The number of filter taps.
cutoff_ratio (float): Cut-off frequency ratio.
beta (float): Beta coefficient for kaiser window.
"""
super().__init__()
......@@ -110,28 +102,20 @@ class PQMF(nn.Layer):
def analysis(self, x):
"""Analysis with PQMF.
Parameters
----------
x : Tensor
Input tensor (B, 1, T).
Returns
----------
Tensor
Output tensor (B, subbands, T // subbands).
Args:
x (Tensor): Input tensor (B, 1, T).
Returns:
Tensor: Output tensor (B, subbands, T // subbands).
"""
x = F.conv1d(self.pad_fn(x), self.analysis_filter)
return F.conv1d(x, self.updown_filter, stride=self.subbands)
def synthesis(self, x):
"""Synthesis with PQMF.
Parameters
----------
x : Tensor
Input tensor (B, subbands, T // subbands).
Returns
----------
Tensor
Output tensor (B, 1, T).
Args:
x (Tensor): Input tensor (B, subbands, T // subbands).
Returns:
Tensor: Output tensor (B, 1, T).
"""
x = F.conv1d_transpose(
x, self.updown_filter * self.subbands, stride=self.subbands)
......
paddlespeech/t2s/modules/predictor/duration_predictor.py
浏览文件 @ 9699c007
......@@ -49,20 +49,13 @@ class DurationPredictor(nn.Layer):
offset=1.0):
"""Initilize duration predictor module.
Parameters
----------
idim : int
Input dimension.
n_layers : int, optional
Number of convolutional layers.
n_chans : int, optional
Number of channels of convolutional layers.
kernel_size : int, optional
Kernel size of convolutional layers.
dropout_rate : float, optional
Dropout rate.
offset : float, optional
Offset value to avoid nan in log domain.
Args:
idim (int):Input dimension.
n_layers (int, optional): Number of convolutional layers.
n_chans (int, optional): Number of channels of convolutional layers.
kernel_size (int, optional): Kernel size of convolutional layers.
dropout_rate (float, optional): Dropout rate.
offset (float, optional): Offset value to avoid nan in log domain.
"""
super().__init__()
......@@ -105,35 +98,23 @@ class DurationPredictor(nn.Layer):
def forward(self, xs, x_masks=None):
"""Calculate forward propagation.
Args:
xs(Tensor): Batch of input sequences (B, Tmax, idim).
x_masks(ByteTensor, optional, optional): Batch of masks indicating padded part (B, Tmax). (Default value = None)
Parameters
----------
xs : Tensor
Batch of input sequences (B, Tmax, idim).
x_masks : ByteTensor, optional
Batch of masks indicating padded part (B, Tmax).
Returns
----------
Tensor
Batch of predicted durations in log domain (B, Tmax).
Returns:
Tensor: Batch of predicted durations in log domain (B, Tmax).
"""
return self._forward(xs, x_masks, False)
def inference(self, xs, x_masks=None):
"""Inference duration.
Args:
xs(Tensor): Batch of input sequences (B, Tmax, idim).
x_masks(Tensor(bool), optional, optional): Batch of masks indicating padded part (B, Tmax). (Default value = None)
Parameters
----------
xs : Tensor
Batch of input sequences (B, Tmax, idim).
x_masks : Tensor(bool), optional
Batch of masks indicating padded part (B, Tmax).
Returns
----------
Tensor
Batch of predicted durations in linear domain int64 (B, Tmax).
Returns:
Tensor: Batch of predicted durations in linear domain int64 (B, Tmax).
"""
return self._forward(xs, x_masks, True)
......@@ -147,13 +128,9 @@ class DurationPredictorLoss(nn.Layer):
def __init__(self, offset=1.0, reduction="mean"):
"""Initilize duration predictor loss module.
Parameters
----------
offset : float, optional
Offset value to avoid nan in log domain.
reduction : str
Reduction type in loss calculation.
Args:
offset (float, optional): Offset value to avoid nan in log domain.
reduction (str): Reduction type in loss calculation.
"""
super().__init__()
self.criterion = nn.MSELoss(reduction=reduction)
......@@ -162,21 +139,15 @@ class DurationPredictorLoss(nn.Layer):
def forward(self, outputs, targets):
"""Calculate forward propagation.
Parameters
----------
outputs : Tensor
Batch of prediction durations in log domain (B, T)
targets : Tensor
Batch of groundtruth durations in linear domain (B, T)
Returns
----------
Tensor
Mean squared error loss value.
Note
----------
`outputs` is in log domain but `targets` is in linear domain.
Args:
outputs(Tensor): Batch of prediction durations in log domain (B, T)
targets(Tensor): Batch of groundtruth durations in linear domain (B, T)
Returns:
Tensor: Mean squared error loss value.
Note:
`outputs` is in log domain but `targets` is in linear domain.
"""
# NOTE: outputs is in log domain while targets in linear
targets = paddle.log(targets.cast(dtype='float32') + self.offset)
......
paddlespeech/t2s/modules/predictor/length_regulator.py
浏览文件 @ 9699c007
......@@ -35,10 +35,8 @@ class LengthRegulator(nn.Layer):
def __init__(self, pad_value=0.0):
"""Initilize length regulator module.
Parameters
----------
pad_value : float, optional
Value used for padding.
Args:
pad_value (float, optional): Value used for padding.
"""
super().__init__()
......@@ -90,19 +88,13 @@ class LengthRegulator(nn.Layer):
def forward(self, xs, ds, alpha=1.0, is_inference=False):
"""Calculate forward propagation.
Parameters
----------
xs : Tensor
Batch of sequences of char or phoneme embeddings (B, Tmax, D).
ds : Tensor(int64)
Batch of durations of each frame (B, T).
alpha : float, optional
Alpha value to control speed of speech.
Args:
xs (Tensor): Batch of sequences of char or phoneme embeddings (B, Tmax, D).
ds (Tensor(int64)): Batch of durations of each frame (B, T).
alpha (float, optional): Alpha value to control speed of speech.
Returns
----------
Tensor
replicated input tensor based on durations (B, T*, D).
Returns:
Tensor: replicated input tensor based on durations (B, T*, D).
"""
if alpha != 1.0:
......
paddlespeech/t2s/modules/predictor/variance_predictor.py
浏览文件 @ 9699c007
......@@ -42,18 +42,12 @@ class VariancePredictor(nn.Layer):
dropout_rate: float=0.5, ):
"""Initilize duration predictor module.
Parameters
----------
idim : int
Input dimension.
n_layers : int, optional
Number of convolutional layers.
n_chans : int, optional
Number of channels of convolutional layers.
kernel_size : int, optional
Kernel size of convolutional layers.
dropout_rate : float, optional
Dropout rate.
Args:
idim (int): Input dimension.
n_layers (int, optional): Number of convolutional layers.
n_chans (int, optional): Number of channels of convolutional layers.
kernel_size (int, optional): Kernel size of convolutional layers.
dropout_rate (float, optional): Dropout rate.
"""
assert check_argument_types()
super().__init__()
......@@ -79,17 +73,12 @@ class VariancePredictor(nn.Layer):
x_masks: paddle.Tensor=None) -> paddle.Tensor:
"""Calculate forward propagation.
Parameters
----------
xs : Tensor
Batch of input sequences (B, Tmax, idim).
x_masks : Tensor(bool), optional
Batch of masks indicating padded part (B, Tmax, 1).
Args:
xs (Tensor): Batch of input sequences (B, Tmax, idim).
x_masks (Tensor(bool), optional): Batch of masks indicating padded part (B, Tmax, 1).
Returns
----------
Tensor
Batch of predicted sequences (B, Tmax, 1).
Returns:
Tensor: Batch of predicted sequences (B, Tmax, 1).
"""
# (B, idim, Tmax)
xs = xs.transpose([0, 2, 1])
......
paddlespeech/t2s/modules/residual_block.py
浏览文件 @ 9699c007
......@@ -28,26 +28,16 @@ class WaveNetResidualBlock(nn.Layer):
unit and parametric redidual and skip connections. For more details,
refer to `WaveNet: A Generative Model for Raw Audio <https://arxiv.org/abs/1609.03499>`_.
Parameters
----------
kernel_size : int, optional
Kernel size of the 1D convolution, by default 3
residual_channels : int, optional
Feature size of the resiaudl output(and also the input), by default 64
gate_channels : int, optional
Output feature size of the 1D convolution, by default 128
skip_channels : int, optional
Feature size of the skip output, by default 64
aux_channels : int, optional
Feature size of the auxiliary input (e.g. spectrogram), by default 80
dropout : float, optional
Probability of the dropout before the 1D convolution, by default 0.
dilation : int, optional
Dilation of the 1D convolution, by default 1
bias : bool, optional
Whether to use bias in the 1D convolution, by default True
use_causal_conv : bool, optional
Whether to use causal padding for the 1D convolution, by default False
Args:
kernel_size (int, optional): Kernel size of the 1D convolution, by default 3
residual_channels (int, optional): Feature size of the resiaudl output(and also the input), by default 64
gate_channels (int, optional): Output feature size of the 1D convolution, by default 128
skip_channels (int, optional): Feature size of the skip output, by default 64
aux_channels (int, optional): Feature size of the auxiliary input (e.g. spectrogram), by default 80
dropout (float, optional): Probability of the dropout before the 1D convolution, by default 0.
dilation (int, optional): Dilation of the 1D convolution, by default 1
bias (bool, optional): Whether to use bias in the 1D convolution, by default True
use_causal_conv (bool, optional): Whether to use causal padding for the 1D convolution, by default False
"""
def __init__(self,
......@@ -90,21 +80,15 @@ class WaveNetResidualBlock(nn.Layer):
def forward(self, x, c):
"""
Parameters
----------
x : Tensor
Shape (N, C_res, T), the input features.
c : Tensor
Shape (N, C_aux, T), the auxiliary input.
Returns
-------
res : Tensor
Shape (N, C_res, T), the residual output, which is used as the
input of the next ResidualBlock in a stack of ResidualBlocks.
skip : Tensor
Shape (N, C_skip, T), the skip output, which is collected among
each layer in a stack of ResidualBlocks.
Args:
x (Tensor): the input features. Shape (N, C_res, T)
c (Tensor): the auxiliary input. Shape (N, C_aux, T)
Returns:
res (Tensor): Shape (N, C_res, T), the residual output, which is used as the
input of the next ResidualBlock in a stack of ResidualBlocks.
skip (Tensor): Shape (N, C_skip, T), the skip output, which is collected among
each layer in a stack of ResidualBlocks.
"""
x_input = x
x = F.dropout(x, self.dropout, training=self.training)
......@@ -136,22 +120,14 @@ class HiFiGANResidualBlock(nn.Layer):
nonlinear_activation_params: Dict[str, Any]={"negative_slope": 0.1},
):
"""Initialize HiFiGANResidualBlock module.
Parameters
----------
kernel_size : int
Kernel size of dilation convolution layer.
channels : int
Number of channels for convolution layer.
dilations : List[int]
List of dilation factors.
use_additional_convs : bool
Whether to use additional convolution layers.
bias : bool
Whether to add bias parameter in convolution layers.
nonlinear_activation : str
Activation function module name.
nonlinear_activation_params : dict
Hyperparameters for activation function.
Args:
kernel_size (int): Kernel size of dilation convolution layer.
channels (int): Number of channels for convolution layer.
dilations (List[int]): List of dilation factors.
use_additional_convs (bool): Whether to use additional convolution layers.
bias (bool): Whether to add bias parameter in convolution layers.
nonlinear_activation (str): Activation function module name.
nonlinear_activation_params (dict): Hyperparameters for activation function.
"""
super().__init__()
......@@ -190,14 +166,10 @@ class HiFiGANResidualBlock(nn.Layer):
def forward(self, x):
"""Calculate forward propagation.
Parameters
----------
x : Tensor
Input tensor (B, channels, T).
Returns
----------
Tensor
Output tensor (B, channels, T).
Args:
x (Tensor): Input tensor (B, channels, T).
Returns:
Tensor: Output tensor (B, channels, T).
"""
for idx in range(len(self.convs1)):
xt = self.convs1[idx](x)
......
paddlespeech/t2s/utils/h5_utils.py
浏览文件 @ 9699c007
此差异已折叠。
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或