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
未验证 提交 e7916a5a 编写于 作者: 小湉湉's avatar 小湉湉 提交者： GitHub
浏览文件
操作

Merge pull request #1582 from KPatr1ck/docs 

[Audio][Doc]Add paddleaudio doc.
上级 e483fa59 831cadac
展开全部 隐藏空白更改
内联 并排
Showing with 550 addition and 544 deletion
paddleaudio/paddleaudio/backends/soundfile_backend.py
浏览文件 @ e7916a5a
......@@ -11,6 +11,7 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import warnings
from typing import Optional
from typing import Tuple
......@@ -19,7 +20,6 @@ from typing import Union
import numpy as np
import resampy
import soundfile as sf
from numpy import ndarray as array
from scipy.io import wavfile
from ..utils import ParameterError
......@@ -38,13 +38,21 @@ RESAMPLE_MODES = ['kaiser_best', 'kaiser_fast']
EPS = 1e-8
def resample(y: array, src_sr: int, target_sr: int,
mode: str='kaiser_fast') -> array:
""" Audio resampling
This function is the same as using resampy.resample().
Notes:
The default mode is kaiser_fast. For better audio quality, use mode = 'kaiser_fast'
"""
def resample(y: np.ndarray,
src_sr: int,
target_sr: int,
mode: str='kaiser_fast') -> np.ndarray:
"""Audio resampling.
Args:
y (np.ndarray): Input waveform array in 1D or 2D.
src_sr (int): Source sample rate.
target_sr (int): Target sample rate.
mode (str, optional): The resampling filter to use. Defaults to 'kaiser_fast'.
Returns:
np.ndarray: `y` resampled to `target_sr`
"""
if mode == 'kaiser_best':
warnings.warn(
......@@ -53,7 +61,7 @@ def resample(y: array, src_sr: int, target_sr: int,
if not isinstance(y, np.ndarray):
raise ParameterError(
'Only support numpy array, but received y in {type(y)}')
'Only support numpy np.ndarray, but received y in {type(y)}')
if mode not in RESAMPLE_MODES:
raise ParameterError(f'resample mode must in {RESAMPLE_MODES}')
......@@ -61,9 +69,17 @@ def resample(y: array, src_sr: int, target_sr: int,
return resampy.resample(y, src_sr, target_sr, filter=mode)
def to_mono(y: array, merge_type: str='average') -> array:
""" convert sterior audio to mono
def to_mono(y: np.ndarray, merge_type: str='average') -> np.ndarray:
"""Convert sterior audio to mono.
Args:
y (np.ndarray): Input waveform array in 1D or 2D.
merge_type (str, optional): Merge type to generate mono waveform. Defaults to 'average'.
Returns:
np.ndarray: `y` with mono channel.
"""
if merge_type not in MERGE_TYPES:
raise ParameterError(
f'Unsupported merge type {merge_type}, available types are {MERGE_TYPES}'
......@@ -101,18 +117,34 @@ def to_mono(y: array, merge_type: str='average') -> array:
return y_out
def _safe_cast(y: array, dtype: Union[type, str]) -> array:
""" data type casting in a safe way, i.e., prevent overflow or underflow
This function is used internally.
def _safe_cast(y: np.ndarray, dtype: Union[type, str]) -> np.ndarray:
"""Data type casting in a safe way, i.e., prevent overflow or underflow.
Args:
y (np.ndarray): Input waveform array in 1D or 2D.
dtype (Union[type, str]): Data type of waveform.
Returns:
np.ndarray: `y` after safe casting.
"""
return np.clip(y, np.iinfo(dtype).min, np.iinfo(dtype).max).astype(dtype)
if 'float' in str(y.dtype):
return np.clip(y, np.finfo(dtype).min,
np.finfo(dtype).max).astype(dtype)
else:
return np.clip(y, np.iinfo(dtype).min,
np.iinfo(dtype).max).astype(dtype)
def depth_convert(y: array, dtype: Union[type, str],
dithering: bool=True) -> array:
"""Convert audio array to target dtype safely
This function convert audio waveform to a target dtype, with addition steps of
def depth_convert(y: np.ndarray, dtype: Union[type, str]) -> np.ndarray:
"""Convert audio array to target dtype safely. This function convert audio waveform to a target dtype, with addition steps of
preventing overflow/underflow and preserving audio range.
Args:
y (np.ndarray): Input waveform array in 1D or 2D.
dtype (Union[type, str]): Data type of waveform.
Returns:
np.ndarray: `y` after safe casting.
"""
SUPPORT_DTYPE = ['int16', 'int8', 'float32', 'float64']
......@@ -157,14 +189,20 @@ def depth_convert(y: array, dtype: Union[type, str],
return y
def sound_file_load(file: str,
def sound_file_load(file: os.PathLike,
offset: Optional[float]=None,
dtype: str='int16',
duration: Optional[int]=None) -> Tuple[array, int]:
"""Load audio using soundfile library
This function load audio file using libsndfile.
Reference:
http://www.mega-nerd.com/libsndfile/#Features
duration: Optional[int]=None) -> Tuple[np.ndarray, int]:
"""Load audio using soundfile library. This function load audio file using libsndfile.
Args:
file (os.PathLike): File of waveform.
offset (Optional[float], optional): Offset to the start of waveform. Defaults to None.
dtype (str, optional): Data type of waveform. Defaults to 'int16'.
duration (Optional[int], optional): Duration of waveform to read. Defaults to None.
Returns:
Tuple[np.ndarray, int]: Waveform in ndarray and its samplerate.
"""
with sf.SoundFile(file) as sf_desc:
sr_native = sf_desc.samplerate
......@@ -179,9 +217,17 @@ def sound_file_load(file: str,
return y, sf_desc.samplerate
def normalize(y: array, norm_type: str='linear',
mul_factor: float=1.0) -> array:
""" normalize an input audio with additional multiplier.
def normalize(y: np.ndarray, norm_type: str='linear',
mul_factor: float=1.0) -> np.ndarray:
"""Normalize an input audio with additional multiplier.
Args:
y (np.ndarray): Input waveform array in 1D or 2D.
norm_type (str, optional): Type of normalization. Defaults to 'linear'.
mul_factor (float, optional): Scaling factor. Defaults to 1.0.
Returns:
np.ndarray: `y` after normalization.
"""
if norm_type == 'linear':
......@@ -199,12 +245,13 @@ def normalize(y: array, norm_type: str='linear',
return y
def save(y: array, sr: int, file: str) -> None:
"""Save audio file to disk.
This function saves audio to disk using scipy.io.wavfile, with additional step
to convert input waveform to int16 unless it already is int16
Notes:
It only support raw wav format.
def save(y: np.ndarray, sr: int, file: os.PathLike) -> None:
"""Save audio file to disk. This function saves audio to disk using scipy.io.wavfile, with additional step to convert input waveform to int16.
Args:
y (np.ndarray): Input waveform array in 1D or 2D.
sr (int): Sample rate.
file (os.PathLike): Path of auido file to save.
"""
if not file.endswith('.wav'):
raise ParameterError(
......@@ -226,7 +273,7 @@ def save(y: array, sr: int, file: str) -> None:
def load(
file: str,
file: os.PathLike,
sr: Optional[int]=None,
mono: bool=True,
merge_type: str='average', # ch0,ch1,random,average
......@@ -236,11 +283,24 @@ def load(
offset: float=0.0,
duration: Optional[int]=None,
dtype: str='float32',
resample_mode: str='kaiser_fast') -> Tuple[array, int]:
"""Load audio file from disk.
This function loads audio from disk using using audio beackend.
Parameters:
Notes:
resample_mode: str='kaiser_fast') -> Tuple[np.ndarray, int]:
"""Load audio file from disk. This function loads audio from disk using using audio beackend.
Args:
file (os.PathLike): Path of auido file to load.
sr (Optional[int], optional): Sample rate of loaded waveform. Defaults to None.
mono (bool, optional): Return waveform with mono channel. Defaults to True.
merge_type (str, optional): Merge type of multi-channels waveform. Defaults to 'average'.
normal (bool, optional): Waveform normalization. Defaults to True.
norm_type (str, optional): Type of normalization. Defaults to 'linear'.
norm_mul_factor (float, optional): Scaling factor. Defaults to 1.0.
offset (float, optional): Offset to the start of waveform. Defaults to 0.0.
duration (Optional[int], optional): Duration of waveform to read. Defaults to None.
dtype (str, optional): Data type of waveform. Defaults to 'float32'.
resample_mode (str, optional): The resampling filter to use. Defaults to 'kaiser_fast'.
Returns:
Tuple[np.ndarray, int]: Waveform in ndarray and its samplerate.
"""
y, r = sound_file_load(file, offset=offset, dtype=dtype, duration=duration)
......
paddleaudio/paddleaudio/compliance/kaldi.py
浏览文件 @ e7916a5a
......@@ -220,7 +220,7 @@ def spectrogram(waveform: Tensor,
"""Compute and return a spectrogram from a waveform. The output is identical to Kaldi's.
Args:
waveform (Tensor): A waveform tensor with shape [C, T].
waveform (Tensor): A waveform tensor with shape `(C, T)`.
blackman_coeff (float, optional): Coefficient for Blackman window.. Defaults to 0.42.
channel (int, optional): Select the channel of waveform. Defaults to -1.
dither (float, optional): Dithering constant . Defaults to 0.0.
......@@ -239,7 +239,7 @@ def spectrogram(waveform: Tensor,
window_type (str, optional): Choose type of window for FFT computation. Defaults to POVEY.
Returns:
Tensor: A spectrogram tensor with shape (m, padded_window_size // 2 + 1) where m is the number of frames
Tensor: A spectrogram tensor with shape `(m, padded_window_size // 2 + 1)` where m is the number of frames
depends on frame_length and frame_shift.
"""
dtype = waveform.dtype
......@@ -422,7 +422,7 @@ def fbank(waveform: Tensor,
"""Compute and return filter banks from a waveform. The output is identical to Kaldi's.
Args:
waveform (Tensor): A waveform tensor with shape [C, T].
waveform (Tensor): A waveform tensor with shape `(C, T)`.
blackman_coeff (float, optional): Coefficient for Blackman window.. Defaults to 0.42.
channel (int, optional): Select the channel of waveform. Defaults to -1.
dither (float, optional): Dithering constant . Defaults to 0.0.
......@@ -451,7 +451,7 @@ def fbank(waveform: Tensor,
window_type (str, optional): Choose type of window for FFT computation. Defaults to POVEY.
Returns:
Tensor: A filter banks tensor with shape (m, n_mels).
Tensor: A filter banks tensor with shape `(m, n_mels)`.
"""
dtype = waveform.dtype
......@@ -542,7 +542,7 @@ def mfcc(waveform: Tensor,
identical to Kaldi's.
Args:
waveform (Tensor): A waveform tensor with shape [C, T].
waveform (Tensor): A waveform tensor with shape `(C, T)`.
blackman_coeff (float, optional): Coefficient for Blackman window.. Defaults to 0.42.
cepstral_lifter (float, optional): Scaling of output mfccs. Defaults to 22.0.
channel (int, optional): Select the channel of waveform. Defaults to -1.
......@@ -571,7 +571,7 @@ def mfcc(waveform: Tensor,
window_type (str, optional): Choose type of window for FFT computation. Defaults to POVEY.
Returns:
Tensor: A mel frequency cepstral coefficients tensor with shape (m, n_mfcc).
Tensor: A mel frequency cepstral coefficients tensor with shape `(m, n_mfcc)`.
"""
assert n_mfcc <= n_mels, 'n_mfcc cannot be larger than n_mels: %d vs %d' % (
n_mfcc, n_mels)
......
paddleaudio/paddleaudio/functional/functional.py
浏览文件 @ e7916a5a
......@@ -17,6 +17,7 @@ from typing import Optional
from typing import Union
import paddle
from paddle import Tensor
__all__ = [
'hz_to_mel',
......@@ -29,19 +30,20 @@ __all__ = [
]
def hz_to_mel(freq: Union[paddle.Tensor, float],
htk: bool=False) -> Union[paddle.Tensor, float]:
def hz_to_mel(freq: Union[Tensor, float],
htk: bool=False) -> Union[Tensor, float]:
"""Convert Hz to Mels.
Parameters:
freq: the input tensor of arbitrary shape, or a single floating point number.
htk: use HTK formula to do the conversion.
The default value is False.
Args:
freq (Union[Tensor, float]): The input tensor with arbitrary shape.
htk (bool, optional): Use htk scaling. Defaults to False.
Returns:
The frequencies represented in Mel-scale.
Union[Tensor, float]: Frequency in mels.
"""
if htk:
if isinstance(freq, paddle.Tensor):
if isinstance(freq, Tensor):
return 2595.0 * paddle.log10(1.0 + freq / 700.0)
else:
return 2595.0 * math.log10(1.0 + freq / 700.0)
......@@ -58,7 +60,7 @@ def hz_to_mel(freq: Union[paddle.Tensor, float],
min_log_mel = (min_log_hz - f_min) / f_sp # same (Mels)
logstep = math.log(6.4) / 27.0 # step size for log region
if isinstance(freq, paddle.Tensor):
if isinstance(freq, Tensor):
target = min_log_mel + paddle.log(
freq / min_log_hz + 1e-10) / logstep # prevent nan with 1e-10
mask = (freq > min_log_hz).astype(freq.dtype)
......@@ -71,14 +73,16 @@ def hz_to_mel(freq: Union[paddle.Tensor, float],
return mels
def mel_to_hz(mel: Union[float, paddle.Tensor],
htk: bool=False) -> Union[float, paddle.Tensor]:
def mel_to_hz(mel: Union[float, Tensor],
htk: bool=False) -> Union[float, Tensor]:
"""Convert mel bin numbers to frequencies.
Parameters:
mel: the mel frequency represented as a tensor of arbitrary shape, or a floating point number.
htk: use HTK formula to do the conversion.
Args:
mel (Union[float, Tensor]): The mel frequency represented as a tensor with arbitrary shape.
htk (bool, optional): Use htk scaling. Defaults to False.
Returns:
The frequencies represented in hz.
Union[float, Tensor]: Frequencies in Hz.
"""
if htk:
return 700.0 * (10.0**(mel / 2595.0) - 1.0)
......@@ -90,7 +94,7 @@ def mel_to_hz(mel: Union[float, paddle.Tensor],
min_log_hz = 1000.0 # beginning of log region (Hz)
min_log_mel = (min_log_hz - f_min) / f_sp # same (Mels)
logstep = math.log(6.4) / 27.0 # step size for log region
if isinstance(mel, paddle.Tensor):
if isinstance(mel, Tensor):
target = min_log_hz * paddle.exp(logstep * (mel - min_log_mel))
mask = (mel > min_log_mel).astype(mel.dtype)
freqs = target * mask + freqs * (
......@@ -106,16 +110,18 @@ def mel_frequencies(n_mels: int=64,
f_min: float=0.0,
f_max: float=11025.0,
htk: bool=False,
dtype: str=paddle.float32):
dtype: str='float32') -> Tensor:
"""Compute mel frequencies.
Parameters:
n_mels(int): number of Mel bins.
f_min(float): the lower cut-off frequency, below which the filter response is zero.
f_max(float): the upper cut-off frequency, above which the filter response is zero.
htk(bool): whether to use htk formula.
dtype(str): the datatype of the return frequencies.
Args:
n_mels (int, optional): Number of mel bins. Defaults to 64.
f_min (float, optional): Minimum frequency in Hz. Defaults to 0.0.
fmax (float, optional): Maximum frequency in Hz. Defaults to 11025.0.
htk (bool, optional): Use htk scaling. Defaults to False.
dtype (str, optional): The data type of the return frequencies. Defaults to 'float32'.
Returns:
The frequencies represented in Mel-scale
Tensor: Tensor of n_mels frequencies in Hz with shape `(n_mels,)`.
"""
# 'Center freqs' of mel bands - uniformly spaced between limits
min_mel = hz_to_mel(f_min, htk=htk)
......@@ -125,14 +131,16 @@ def mel_frequencies(n_mels: int=64,
return freqs
def fft_frequencies(sr: int, n_fft: int, dtype: str=paddle.float32):
def fft_frequencies(sr: int, n_fft: int, dtype: str='float32') -> Tensor:
"""Compute fourier frequencies.
Parameters:
sr(int): the audio sample rate.
n_fft(float): the number of fft bins.
dtype(str): the datatype of the return frequencies.
Args:
sr (int): Sample rate.
n_fft (int): Number of fft bins.
dtype (str, optional): The data type of the return frequencies. Defaults to 'float32'.
Returns:
The frequencies represented in hz.
Tensor: FFT frequencies in Hz with shape `(n_fft//2 + 1,)`.
"""
return paddle.linspace(0, float(sr) / 2, int(1 + n_fft // 2), dtype=dtype)
......@@ -144,23 +152,21 @@ def compute_fbank_matrix(sr: int,
f_max: Optional[float]=None,
htk: bool=False,
norm: Union[str, float]='slaney',
dtype: str=paddle.float32):
dtype: str='float32') -> Tensor:
"""Compute fbank matrix.
Parameters:
sr(int): the audio sample rate.
n_fft(int): the number of fft bins.
n_mels(int): the number of Mel bins.
f_min(float): the lower cut-off frequency, below which the filter response is zero.
f_max(float): the upper cut-off frequency, above which the filter response is zero.
htk: whether to use htk formula.
return_complex(bool): whether to return complex matrix. If True, the matrix will
be complex type. Otherwise, the real and image part will be stored in the last
axis of returned tensor.
dtype(str): the datatype of the returned fbank matrix.
Args:
sr (int): Sample rate.
n_fft (int): Number of fft bins.
n_mels (int, optional): Number of mel bins. Defaults to 64.
f_min (float, optional): Minimum frequency in Hz. Defaults to 0.0.
f_max (Optional[float], optional): Maximum frequency in Hz. Defaults to None.
htk (bool, optional): Use htk scaling. Defaults to False.
norm (Union[str, float], optional): Type of normalization. Defaults to 'slaney'.
dtype (str, optional): The data type of the return matrix. Defaults to 'float32'.
Returns:
The fbank matrix of shape (n_mels, int(1+n_fft//2)).
Shape:
output: (n_mels, int(1+n_fft//2))
Tensor: Mel transform matrix with shape `(n_mels, n_fft//2 + 1)`.
"""
if f_max is None:
......@@ -199,27 +205,20 @@ def compute_fbank_matrix(sr: int,
return weights
def power_to_db(magnitude: paddle.Tensor,
def power_to_db(spect: Tensor,
ref_value: float=1.0,
amin: float=1e-10,
top_db: Optional[float]=None) -> paddle.Tensor:
"""Convert a power spectrogram (amplitude squared) to decibel (dB) units.
The function computes the scaling ``10 * log10(x / ref)`` in a numerically
stable way.
Parameters:
magnitude(Tensor): the input magnitude tensor of any shape.
ref_value(float): the reference value. If smaller than 1.0, the db level
of the signal will be pulled up accordingly. Otherwise, the db level
is pushed down.
amin(float): the minimum value of input magnitude, below which the input
magnitude is clipped(to amin).
top_db(float): the maximum db value of resulting spectrum, above which the
spectrum is clipped(to top_db).
top_db: Optional[float]=None) -> Tensor:
"""Convert a power spectrogram (amplitude squared) to decibel (dB) units. The function computes the scaling `10 * log10(x / ref)` in a numerically stable way.
Args:
spect (Tensor): STFT power spectrogram.
ref_value (float, optional): The reference value. If smaller than 1.0, the db level of the signal will be pulled up accordingly. Otherwise, the db level is pushed down. Defaults to 1.0.
amin (float, optional): Minimum threshold. Defaults to 1e-10.
top_db (Optional[float], optional): Threshold the output at `top_db` below the peak. Defaults to None.
Returns:
The spectrogram in log-scale.
shape:
input: any shape
output: same as input
Tensor: Power spectrogram in db scale.
"""
if amin <= 0:
raise Exception("amin must be strictly positive")
......@@ -227,8 +226,8 @@ def power_to_db(magnitude: paddle.Tensor,
if ref_value <= 0:
raise Exception("ref_value must be strictly positive")
ones = paddle.ones_like(magnitude)
log_spec = 10.0 * paddle.log10(paddle.maximum(ones * amin, magnitude))
ones = paddle.ones_like(spect)
log_spec = 10.0 * paddle.log10(paddle.maximum(ones * amin, spect))
log_spec -= 10.0 * math.log10(max(ref_value, amin))
if top_db is not None:
......@@ -242,15 +241,17 @@ def power_to_db(magnitude: paddle.Tensor,
def create_dct(n_mfcc: int,
n_mels: int,
norm: Optional[str]='ortho',
dtype: Optional[str]=paddle.float32) -> paddle.Tensor:
dtype: str='float32') -> Tensor:
"""Create a discrete cosine transform(DCT) matrix.
Parameters:
Args:
n_mfcc (int): Number of mel frequency cepstral coefficients.
n_mels (int): Number of mel filterbanks.
norm (str, optional): Normalizaiton type. Defaults to 'ortho'.
norm (Optional[str], optional): Normalizaiton type. Defaults to 'ortho'.
dtype (str, optional): The data type of the return matrix. Defaults to 'float32'.
Returns:
Tensor: The DCT matrix with shape (n_mels, n_mfcc).
Tensor: The DCT matrix with shape `(n_mels, n_mfcc)`.
"""
n = paddle.arange(n_mels, dtype=dtype)
k = paddle.arange(n_mfcc, dtype=dtype).unsqueeze(1)
......
paddleaudio/paddleaudio/functional/window.py
浏览文件 @ e7916a5a
......@@ -20,24 +20,11 @@ from paddle import Tensor
__all__ = [
'get_window',
# windows
'taylor',
'hamming',
'hann',
'tukey',
'kaiser',
'gaussian',
'exponential',
'triang',
'bohman',
'blackman',
'cosine',
]
def _cat(a: List[Tensor], data_type: str) -> Tensor:
l = [paddle.to_tensor(_a, data_type) for _a in a]
def _cat(x: List[Tensor], data_type: str) -> Tensor:
l = [paddle.to_tensor(_, data_type) for _ in x]
return paddle.concat(l)
......@@ -48,7 +35,7 @@ def _acosh(x: Union[Tensor, float]) -> Tensor:
def _extend(M: int, sym: bool) -> bool:
"""Extend window by 1 sample if needed for DFT-even symmetry"""
"""Extend window by 1 sample if needed for DFT-even symmetry. """
if not sym:
return M + 1, True
else:
......@@ -56,7 +43,7 @@ def _extend(M: int, sym: bool) -> bool:
def _len_guards(M: int) -> bool:
"""Handle small or incorrect window lengths"""
"""Handle small or incorrect window lengths. """
if int(M) != M or M < 0:
raise ValueError('Window length M must be a non-negative integer')
......@@ -64,15 +51,15 @@ def _len_guards(M: int) -> bool:
def _truncate(w: Tensor, needed: bool) -> Tensor:
"""Truncate window by 1 sample if needed for DFT-even symmetry"""
"""Truncate window by 1 sample if needed for DFT-even symmetry. """
if needed:
return w[:-1]
else:
return w
def general_gaussian(M: int, p, sig, sym: bool=True,
dtype: str='float64') -> Tensor:
def _general_gaussian(M: int, p, sig, sym: bool=True,
dtype: str='float64') -> Tensor:
"""Compute a window with a generalized Gaussian shape.
This function is consistent with scipy.signal.windows.general_gaussian().
"""
......@@ -86,8 +73,8 @@ def general_gaussian(M: int, p, sig, sym: bool=True,
return _truncate(w, needs_trunc)
def general_cosine(M: int, a: float, sym: bool=True,
dtype: str='float64') -> Tensor:
def _general_cosine(M: int, a: float, sym: bool=True,
dtype: str='float64') -> Tensor:
"""Compute a generic weighted sum of cosine terms window.
This function is consistent with scipy.signal.windows.general_cosine().
"""
......@@ -101,31 +88,23 @@ def general_cosine(M: int, a: float, sym: bool=True,
return _truncate(w, needs_trunc)
def general_hamming(M: int, alpha: float, sym: bool=True,
dtype: str='float64') -> Tensor:
def _general_hamming(M: int, alpha: float, sym: bool=True,
dtype: str='float64') -> Tensor:
"""Compute a generalized Hamming window.
This function is consistent with scipy.signal.windows.general_hamming()
"""
return general_cosine(M, [alpha, 1. - alpha], sym, dtype=dtype)
return _general_cosine(M, [alpha, 1. - alpha], sym, dtype=dtype)
def taylor(M: int,
nbar=4,
sll=30,
norm=True,
sym: bool=True,
dtype: str='float64') -> Tensor:
def _taylor(M: int,
nbar=4,
sll=30,
norm=True,
sym: bool=True,
dtype: str='float64') -> Tensor:
"""Compute a Taylor window.
The Taylor window taper function approximates the Dolph-Chebyshev window's
constant sidelobe level for a parameterized number of near-in sidelobes.
Parameters:
M(int): window size
nbar, sil, norm: the window-specific parameter.
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
if _len_guards(M):
return paddle.ones((M, ), dtype=dtype)
......@@ -171,46 +150,25 @@ def taylor(M: int,
return _truncate(w, needs_trunc)
def hamming(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
def _hamming(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
"""Compute a Hamming window.
The Hamming window is a taper formed by using a raised cosine with
non-zero endpoints, optimized to minimize the nearest side lobe.
Parameters:
M(int): window size
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
return general_hamming(M, 0.54, sym, dtype=dtype)
return _general_hamming(M, 0.54, sym, dtype=dtype)
def hann(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
def _hann(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
"""Compute a Hann window.
The Hann window is a taper formed by using a raised cosine or sine-squared
with ends that touch zero.
Parameters:
M(int): window size
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
return general_hamming(M, 0.5, sym, dtype=dtype)
return _general_hamming(M, 0.5, sym, dtype=dtype)
def tukey(M: int, alpha=0.5, sym: bool=True, dtype: str='float64') -> Tensor:
def _tukey(M: int, alpha=0.5, sym: bool=True, dtype: str='float64') -> Tensor:
"""Compute a Tukey window.
The Tukey window is also known as a tapered cosine window.
Parameters:
M(int): window size
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
if _len_guards(M):
return paddle.ones((M, ), dtype=dtype)
......@@ -237,32 +195,18 @@ def tukey(M: int, alpha=0.5, sym: bool=True, dtype: str='float64') -> Tensor:
return _truncate(w, needs_trunc)
def kaiser(M: int, beta: float, sym: bool=True, dtype: str='float64') -> Tensor:
def _kaiser(M: int, beta: float, sym: bool=True,
dtype: str='float64') -> Tensor:
"""Compute a Kaiser window.
The Kaiser window is a taper formed by using a Bessel function.
Parameters:
M(int): window size.
beta(float): the window-specific parameter.
sym(bool)：whether to return symmetric window.
The default value is True
Returns:
Tensor: the window tensor
"""
raise NotImplementedError()
def gaussian(M: int, std: float, sym: bool=True,
dtype: str='float64') -> Tensor:
def _gaussian(M: int, std: float, sym: bool=True,
dtype: str='float64') -> Tensor:
"""Compute a Gaussian window.
The Gaussian widows has a Gaussian shape defined by the standard deviation(std).
Parameters:
M(int): window size.
std(float): the window-specific parameter.
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
if _len_guards(M):
return paddle.ones((M, ), dtype=dtype)
......@@ -275,21 +219,12 @@ def gaussian(M: int, std: float, sym: bool=True,
return _truncate(w, needs_trunc)
def exponential(M: int,
center=None,
tau=1.,
sym: bool=True,
dtype: str='float64') -> Tensor:
"""Compute an exponential (or Poisson) window.
Parameters:
M(int): window size.
tau(float): the window-specific parameter.
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
def _exponential(M: int,
center=None,
tau=1.,
sym: bool=True,
dtype: str='float64') -> Tensor:
"""Compute an exponential (or Poisson) window. """
if sym and center is not None:
raise ValueError("If sym==True, center must be None.")
if _len_guards(M):
......@@ -305,15 +240,8 @@ def exponential(M: int,
return _truncate(w, needs_trunc)
def triang(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
def _triang(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
"""Compute a triangular window.
Parameters:
M(int): window size.
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
if _len_guards(M):
return paddle.ones((M, ), dtype=dtype)
......@@ -330,16 +258,9 @@ def triang(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
return _truncate(w, needs_trunc)
def bohman(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
def _bohman(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
"""Compute a Bohman window.
The Bohman window is the autocorrelation of a cosine window.
Parameters:
M(int): window size.
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
if _len_guards(M):
return paddle.ones((M, ), dtype=dtype)
......@@ -353,32 +274,18 @@ def bohman(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
return _truncate(w, needs_trunc)
def blackman(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
def _blackman(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
"""Compute a Blackman window.
The Blackman window is a taper formed by using the first three terms of
a summation of cosines. It was designed to have close to the minimal
leakage possible. It is close to optimal, only slightly worse than a
Kaiser window.
Parameters:
M(int): window size.
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
return general_cosine(M, [0.42, 0.50, 0.08], sym, dtype=dtype)
return _general_cosine(M, [0.42, 0.50, 0.08], sym, dtype=dtype)
def cosine(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
def _cosine(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
"""Compute a window with a simple cosine shape.
Parameters:
M(int): window size.
sym(bool)：whether to return symmetric window.
The default value is True
dtype(str): the datatype of returned tensor.
Returns:
Tensor: the window tensor
"""
if _len_guards(M):
return paddle.ones((M, ), dtype=dtype)
......@@ -388,19 +295,20 @@ def cosine(M: int, sym: bool=True, dtype: str='float64') -> Tensor:
return _truncate(w, needs_trunc)
## factory function
def get_window(window: Union[str, Tuple[str, float]],
win_length: int,
fftbins: bool=True,
dtype: str='float64') -> Tensor:
"""Return a window of a given length and type.
Parameters:
window(str|(str,float)): the type of window to create.
win_length(int): the number of samples in the window.
fftbins(bool): If True, create a "periodic" window. Otherwise,
create a "symmetric" window, for use in filter design.
Args:
window (Union[str, Tuple[str, float]]): The window function applied to the signal before the Fourier transform. Supported window functions: 'hamming', 'hann', 'kaiser', 'gaussian', 'exponential', 'triang', 'bohman', 'blackman', 'cosine', 'tukey', 'taylor'.
win_length (int): Number of samples.
fftbins (bool, optional): If True, create a "periodic" window. Otherwise, create a "symmetric" window, for use in filter design. Defaults to True.
dtype (str, optional): The data type of the return window. Defaults to 'float64'.
Returns:
The window represented as a tensor.
Tensor: The window represented as a tensor.
"""
sym = not fftbins
......@@ -420,7 +328,7 @@ def get_window(window: Union[str, Tuple[str, float]],
str(type(window)))
try:
winfunc = eval(winstr)
winfunc = eval('_' + winstr)
except KeyError as e:
raise ValueError("Unknown window type.") from e
......
paddleaudio/paddleaudio/metric/dtw.py
浏览文件 @ e7916a5a
......@@ -20,9 +20,7 @@ __all__ = [
def dtw_distance(xs: np.ndarray, ys: np.ndarray) -> float:
"""dtw distance
Dynamic Time Warping.
"""Dynamic Time Warping.
This function keeps a compact matrix, not the full warping paths matrix.
Uses dynamic programming to compute:
......
paddlespeech/cli/executor.py
浏览文件 @ e7916a5a
......@@ -178,7 +178,8 @@ class BaseExecutor(ABC):
Returns:
bool: return `True` for job input, `False` otherwise.
"""
return input_ and os.path.isfile(input_) and input_.endswith('.job')
return input_ and os.path.isfile(input_) and (input_.endswith('.job') or
input_.endswith('.txt'))
def _get_job_contents(
self, job_input: os.PathLike) -> Dict[str, Union[str, os.PathLike]]:
......
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