refactor data, build vocab; add format data

deepspeech/frontend/utility.py

@@ -29,40 +29,79 @@ logger = logging.getLogger(__name__)
 
 __all__ = [
     "load_cmvn", "read_manifest", "rms_to_db", "rms_to_dbfs", "max_dbfs",
-    "mean_dbfs", "gain_db_to_ratio", "normalize_audio"
+    "mean_dbfs", "gain_db_to_ratio", "normalize_audio", "SOS", "EOS", "UNK",
+    "BLANK"
 ]
 
+IGNORE_ID = -1
+SOS = "<sos/eos>"
+EOS = SOS
+UNK = "<unk>"
+BLANK = "<blank>"
+
+# """Load and parse manifest file.
+
+# Instances with durations outside [min_duration, max_duration] will be
+# filtered out.
+
+# :param manifest_path: Manifest file to load and parse.
+# :type manifest_path: str
+# :param max_duration:maximum output seq length, in seconds for raw wav, in frame numbers for feature data.
+# :type max_duration: float
+# :param min_duration: minimum input seq length, in seconds for raw wav, in frame numbers for feature data.
+# :type min_duration: float
+# :return: Manifest parsing results. List of dict.
+# :rtype: list
+# :raises IOError: If failed to parse the manifest.
+# """
+
+
+def read_manifest(
+        manifest_path,
+        max_input_len=float('inf'),
+        min_input_len=0.0,
+        max_output_len=500.0,
+        min_output_len=0.0,
+        max_output_input_ratio=10.0,
+        min_output_input_ratio=0.05, ):
 
-def read_manifest(manifest_path, max_duration=float('inf'), min_duration=0.0):
-    """Load and parse manifest file.
-
-    Instances with durations outside [min_duration, max_duration] will be
-    filtered out.
-
-    :param manifest_path: Manifest file to load and parse.
-    :type manifest_path: str
-    :param max_duration: Maximal duration in seconds for instance filter.
-    :type max_duration: float
-    :param min_duration: Minimal duration in seconds for instance filter.
-    :type min_duration: float
-    :return: Manifest parsing results. List of dict.
-    :rtype: list
-    :raises IOError: If failed to parse the manifest.
-    """
     manifest = []
     for json_line in codecs.open(manifest_path, 'r', 'utf-8'):
         try:
             json_data = json.loads(json_line)
         except Exception as e:
             raise IOError("Error reading manifest: %s" % str(e))
-        if (json_data["duration"] <= max_duration and
-                json_data["duration"] >= min_duration):
+        feat_len = json_data["feat_shape"][0]
+        token_len = json_data["token_shape"][0]
+        conditions = [
+            feat_len > min_input_len,
+            feat_len < max_input_len,
+            token_len > min_output_len,
+            token_len < max_output_len,
+            token_len / feat_len > min_output_input_ratio,
+            token_len / feat_len < max_output_input_ratio,
+        ]
+        if all(conditions):
             manifest.append(json_data)
     return manifest
 
+    # parser.add_argument('--max_input_len', type=float,
+    #                     default=20,
+    #                     help='maximum output seq length, in seconds for raw wav, in frame numbers for feature data')
+    # parser.add_argument('--min_output_len', type=float,
+    #                     default=0, help='minimum input seq length, in modeling units')
+    # parser.add_argument('--max_output_len', type=float,
+    #                     default=500,
+    #                     help='maximum output seq length, in modeling units')
+    # parser.add_argument('--min_output_input_ratio', type=float, default=0.05,
+    #                     help='minimum output seq length/output seq length ratio')
+    # parser.add_argument('--max_output_input_ratio', type=float, default=10,
+    #                     help='maximum output seq length/output seq length ratio')
+
 
 def rms_to_db(rms: float):
     """Root Mean Square to dB.
+
     Args:
         rms ([float]): root mean square
 
@@ -145,8 +184,10 @@ def normalize_audio(sample_data: np.ndarray, dbfs: float=-3.0103):
 
 def _load_json_cmvn(json_cmvn_file):
     """ Load the json format cmvn stats file and calculate cmvn
+
     Args:
         json_cmvn_file: cmvn stats file in json format
+
     Returns:
         a numpy array of [means, vars]
     """
@@ -168,10 +209,12 @@ def _load_json_cmvn(json_cmvn_file):
 
 def _load_kaldi_cmvn(kaldi_cmvn_file):
     """ Load the kaldi format cmvn stats file and calculate cmvn
+
     Args:
         kaldi_cmvn_file:  kaldi text style global cmvn file, which
            is generated by:
            compute-cmvn-stats --binary=false scp:feats.scp global_cmvn
+
     Returns:
         a numpy array of [means, vars]
     """

deepspeech/io/collator.py

@@ -17,7 +17,7 @@ import numpy as np
 from collections import namedtuple
 
 from deepspeech.io.utility import pad_sequence
-from deepspeech.utils.tensor_utils import IGNORE_ID
+from deepspeech.frontend.utility import IGNORE_ID
 
 logger = logging.getLogger(__name__)
 

deepspeech/models/u2.py

@@ -42,11 +42,11 @@ from deepspeech.modules.decoder import TransformerDecoder
 from deepspeech.modules.loss import LabelSmoothingLoss
 
 from deepspeech.frontend.utility import load_cmvn
+from deepspeech.frontend.utility import IGNORE_ID
 
 from deepspeech.utils import checkpoint
 from deepspeech.utils import layer_tools
 from deepspeech.utils.utility import log_add
-from deepspeech.utils.tensor_utils import IGNORE_ID
 from deepspeech.utils.tensor_utils import add_sos_eos
 from deepspeech.utils.tensor_utils import th_accuracy
 from deepspeech.utils.tensor_utils import pad_sequence

deepspeech/utils/tensor_utils.py

@@ -22,8 +22,6 @@ logger = logging.getLogger(__name__)
 
 __all__ = ["pad_sequence", "add_sos_eos", "th_accuracy"]
 
-IGNORE_ID = -1
-
 
 def pad_sequence(sequences: List[paddle.Tensor],
                  batch_first: bool=False,

examples/dataset/aishell/aishell.py

@@ -62,9 +62,9 @@ def create_manifest(data_dir, manifest_path_prefix):
         transcript_dict[audio_id] = text
 
     data_types = ['train', 'dev', 'test']
-    for type in data_types:
+    for dtype in data_types:
         del json_lines[:]
-        audio_dir = os.path.join(data_dir, 'wav', type)
+        audio_dir = os.path.join(data_dir, 'wav', dtype)
         for subfolder, _, filelist in sorted(os.walk(audio_dir)):
             for fname in filelist:
                 audio_path = os.path.join(subfolder, fname)
@@ -78,12 +78,16 @@ def create_manifest(data_dir, manifest_path_prefix):
                 json_lines.append(
                     json.dumps(
                         {
-                            'audio_filepath': audio_path,
-                            'duration': duration,
-                            'text': text
+                            'utt':
+                            os.path.splitext(os.path.basename(audio_path))[0],
+                            'feat':
+                            audio_path,
+                            'feat_shape': (duration, ),  #second
+                            'text':
+                            text
                         },
                         ensure_ascii=False))
-        manifest_path = manifest_path_prefix + '.' + type
+        manifest_path = manifest_path_prefix + '.' + dtype
         with codecs.open(manifest_path, 'w', 'utf-8') as fout:
             for line in json_lines:
                 fout.write(line + '\n')

examples/dataset/chime3_background/chime3_background.py

@@ -95,10 +95,13 @@ def create_manifest(data_dir, manifest_path):
                 audio_data, samplerate = soundfile.read(filepath)
                 duration = float(len(audio_data)) / samplerate
                 json_lines.append(
-                    json.dumps({
-                        'audio_filepath': filepath,
-                        'duration': duration,
-                        'text': ''
+                    json.dumps(
+                        {
+                            'utt': os.path.splitext(os.path.basename(filepath))[
+                                0],
+                            'feat': filepath,
+                            'feat_shape': (duration, ),  #second
+                            'type': 'background'
                         }))
     with io.open(manifest_path, mode='w', encoding='utf8') as out_file:
         for line in json_lines:

examples/dataset/librispeech/librispeech.py

@@ -89,9 +89,13 @@ def create_manifest(data_dir, manifest_path):
                 duration = float(len(audio_data)) / samplerate
                 json_lines.append(
                     json.dumps({
-                        'audio_filepath': audio_filepath,
-                        'duration': duration,
-                        'text': text
+                        'utt':
+                        os.path.splitext(os.path.basename(audio_filepath))[0],
+                        'feat':
+                        audio_filepath,
+                        'feat_shape': (duration, ),  #second
+                        'text':
+                        text
                     }))
     with codecs.open(manifest_path, 'w', 'utf-8') as out_file:
         for line in json_lines:

examples/dataset/mini_librispeech/mini_librispeech.py

@@ -71,9 +71,13 @@ def create_manifest(data_dir, manifest_path):
                 duration = float(len(audio_data)) / samplerate
                 json_lines.append(
                     json.dumps({
-                        'audio_filepath': audio_filepath,
-                        'duration': duration,
-                        'text': text
+                        'utt':
+                        os.path.splitext(os.path.basename(audio_filepath))[0],
+                        'feat':
+                        audio_filepath,
+                        'feat_shape': (duration, ),  #second
+                        'text':
+                        text
                     }))
     with codecs.open(manifest_path, 'w', 'utf-8') as out_file:
         for line in json_lines:

examples/dataset/musan/musan.py

@@ -53,9 +53,9 @@ def create_manifest(data_dir, manifest_path_prefix):
     print("Creating manifest %s ..." % manifest_path_prefix)
     json_lines = []
     data_types = ['music', 'noise', 'speech']
-    for type in data_types:
+    for dtype in data_types:
         del json_lines[:]
-        audio_dir = os.path.join(data_dir, type)
+        audio_dir = os.path.join(data_dir, dtype)
         for subfolder, _, filelist in sorted(os.walk(audio_dir)):
             print('x, ', subfolder)
             for fname in filelist:
@@ -67,12 +67,16 @@ def create_manifest(data_dir, manifest_path_prefix):
                 json_lines.append(
                     json.dumps(
                         {
-                            'audio_filepath': audio_path,
-                            'duration': duration,
-                            'type': type,
+                            'utt':
+                            os.path.splitext(os.path.basename(audio_path))[0],
+                            'feat':
+                            audio_path,
+                            'feat_shape': (duration, ),  #second
+                            'type':
+                            dtype,
                         },
                         ensure_ascii=False))
-        manifest_path = manifest_path_prefix + '.' + type
+        manifest_path = manifest_path_prefix + '.' + dtype
         with codecs.open(manifest_path, 'w', 'utf-8') as fout:
             for line in json_lines:
                 fout.write(line + '\n')

examples/dataset/rir_noise/rir_noise.py

@@ -55,9 +55,9 @@ def create_manifest(data_dir, manifest_path_prefix):
     data_types = [
         'pointsource_noises', 'real_rirs_isotropic_noises', 'simulated_rirs'
     ]
-    for type in data_types:
+    for dtype in data_types:
         del json_lines[:]
-        audio_dir = os.path.join(data_dir, type)
+        audio_dir = os.path.join(data_dir, dtype)
         for subfolder, _, filelist in sorted(os.walk(audio_dir)):
             for fname in filelist:
                 audio_path = os.path.join(subfolder, fname)
@@ -68,12 +68,16 @@ def create_manifest(data_dir, manifest_path_prefix):
                 json_lines.append(
                     json.dumps(
                         {
-                            'audio_filepath': audio_path,
-                            'duration': duration,
-                            'type': type,
+                            'utt':
+                            os.path.splitext(os.path.basename(audio_path))[0],
+                            'feat':
+                            audio_path,
+                            'feat_shape': (duration, ),  #second
+                            'type':
+                            dtype,
                         },
                         ensure_ascii=False))
-        manifest_path = manifest_path_prefix + '.' + type
+        manifest_path = manifest_path_prefix + '.' + dtype
         with codecs.open(manifest_path, 'w', 'utf-8') as fout:
             for line in json_lines:
                 fout.write(line + '\n')

examples/dataset/voxforge/voxforge.py

@@ -174,8 +174,9 @@ def generate_manifest(data_dir, manifest_path):
             duration = float(len(audio_data)) / samplerate
             json_lines.append(
                 json.dumps({
-                    'audio_filepath': u,
-                    'duration': duration,
+                    'utt': os.path.splitext(os.path.basename(u))[0],
+                    'feat': u,
+                    'feat_shape': (duration, ),  #second
                     'text': trans.lower()
                 }))
 

examples/tiny/s0/local/data.sh

@@ -15,13 +15,20 @@ if [ $? -ne 0 ]; then
     exit 1
 fi
 
-head -n 64 data/manifest.dev-clean  > data/manifest.tiny
+head -n 64 data/manifest.dev-clean  > data/manifest.tiny.raw
 
+# bpemode (unigram or bpe)
+nbpe=200
+bpemode=unigram
+bpeprefix="data/bpe_${bpemode}_${nbpe}"
 # build vocabulary
 python3 ${MAIN_ROOT}/utils/build_vocab.py \
---count_threshold=0 \
+--unit_type "bpe" \
+--count_threshold=${nbpe} \
+--bpe_mode ${bpemode} \
+--bpe_model_prefix ${bpeprefix} \
 --vocab_path="data/vocab.txt" \
---manifest_paths="data/manifest.tiny"
+--manifest_paths="data/manifest.tiny.raw"
 
 if [ $? -ne 0 ]; then
     echo "Build vocabulary failed. Terminated."
@@ -31,7 +38,7 @@ fi
 
 # compute mean and stddev for normalizer
 python3 ${MAIN_ROOT}/utils/compute_mean_std.py \
---manifest_path="data/manifest.tiny" \
+--manifest_path="data/manifest.tiny.raw" \
 --num_samples=64 \
 --specgram_type="linear" \
 --output_path="data/mean_std.npz"
@@ -41,5 +48,21 @@ if [ $? -ne 0 ]; then
     exit 1
 fi
 
+
+# format manifest with tokenids, vocab size
+python3 ${MAIN_ROOT}/utils/format_data.py \
+--feat_type "raw" \
+--unit_type "bpe" \
+--bpe_model_prefix ${bpeprefix} \
+--vocab_path="data/vocab.txt" \
+--manifest_path="data/manifest.tiny.raw" \
+--output_path="data/manifest.tiny"
+
+
+if [ $? -ne 0 ]; then
+    echo "Formt mnaifest failed. Terminated."
+    exit 1
+fi
+
 echo "LibriSpeech Data preparation done."
 exit 0
\ No newline at end of file

utils/build_vocab.py

@@ -17,18 +17,24 @@ Each item in vocabulary file is a character.
 
 import argparse
 import functools
-import codecs
 import json
 from collections import Counter
-import os.path
+import os
+import copy
+import tempfile
 
 from deepspeech.frontend.utility import read_manifest
-from deepspeech.utils.utility import add_arguments, print_arguments
+from deepspeech.frontend.utility import UNK
+from deepspeech.frontend.utility import BLANK
+from deepspeech.frontend.utility import SOS
+from deepspeech.utils.utility import add_arguments
+from deepspeech.utils.utility import print_arguments
 
 parser = argparse.ArgumentParser(description=__doc__)
 add_arg = functools.partial(add_arguments, argparser=parser)
 # yapf: disable
-add_arg('count_threshold',  int,    0,  "Truncation threshold for char counts.")
+add_arg('unit_type', str, "character", "Unit type, e.g. character, word, bpe")
+add_arg('count_threshold',  int,    0,  "Truncation threshold for char/word/bpe counts.")
 add_arg('vocab_path',       str,
         'examples/librispeech/data/vocab.txt',
         "Filepath to write the vocabulary.")
@@ -38,6 +44,11 @@ add_arg('manifest_paths',   str,
         "You can provide multiple manifest files.",
         nargs='+',
         required=True)
+# bpe
+add_arg('bpe_mode', str, 'unigram',
+    "bpe model type, e.g. unigram, bpe, char, word. only need when `unit_type` is bpe")
+add_arg('bpe_model_prefix', str, "bpe_model_%(bpe_mode)_%(count_threshold)",
+    "bpe model prefix, only need when `unit_type` is bpe")
 # yapf: disable
 args = parser.parse_args()
 
@@ -45,23 +56,96 @@ args = parser.parse_args()
 def count_manifest(counter, manifest_path):
     manifest_jsons = read_manifest(manifest_path)
     for line_json in manifest_jsons:
+        if args.unit_type == 'character':
             for char in line_json['text']:
                 counter.update(char)
+        elif args.unit_type == 'word':
+            for word in line_json['text'].split():
+                counter.update(word)
 
+def read_text_manifest(fileobj, manifest_path):
+    manifest_jsons = read_manifest(manifest_path)
+    for line_json in manifest_jsons:
+        fileobj.write(line_json['text'] + "\n")
 
 def main():
     print_arguments(args)
 
+    fout = open(args.vocab_path, 'w', encoding='utf-8')
+    fout.write(BLANK + "\n") # 0 will be used for "blank" in CTC
+    fout.write(UNK + '\n')   # <unk> must be 1
+
+    if args.unit_type != 'bpe':
         counter = Counter()
         for manifest_path in args.manifest_paths:
             count_manifest(counter, manifest_path)
 
         count_sorted = sorted(counter.items(), key=lambda x: x[1], reverse=True)
-    with codecs.open(args.vocab_path, 'w', 'utf-8') as fout:
-        fout.write('<unk>' + '\n')
         for char, count in count_sorted:
             if count < args.count_threshold: break
             fout.write(char + '\n')
+    else:
+        # tools/spm_train --input=$wave_data/lang_char/input.txt 
+        # --vocab_size=${nbpe} --model_type=${bpemode} 
+        # --model_prefix=${bpemodel} --input_sentence_size=100000000
+        import sentencepiece as spm
+
+        fp = tempfile.NamedTemporaryFile(mode='w', delete=False)
+        for manifest_path in args.manifest_paths:
+            read_text_manifest(fp, manifest_path)
+        fp.close()
+        # train
+        spm.SentencePieceTrainer.Train(
+            input=fp.name,
+            vocab_size=args.count_threshold,
+            model_type=args.bpe_mode,
+            model_prefix=args.bpe_model_prefix,
+            input_sentence_size=100000000,
+            character_coverage=0.9995)
+        os.unlink(fp.name)
+
+        # encode
+        sp = spm.SentencePieceProcessor()
+        sp.Load(args.bpe_model_prefix + '.model')
+        stats = {"num_empty": 0, "num_filtered": 0}
+
+        def valid(line):
+            return True
+
+        def encode(l):
+            return sp.EncodeAsPieces(l)
+
+        def encode_line(line):
+            line = line.strip()
+            if len(line) > 0:
+                line = encode(line)
+                if valid(line):
+                    return line
+                else:
+                    stats["num_filtered"] += 1
+            else:
+                stats["num_empty"] += 1
+            return None
+
+        vocabs = set()
+        for manifest_path in args.manifest_paths:
+            manifest_jsons = read_manifest(manifest_path)
+            for line_json in manifest_jsons:
+                line = line_json['text']
+                enc_line = encode_line(line)
+                for code in enc_line:
+                    vocabs.add(code)
+                #print(" ".join(enc_line))
+        vocabs_sorted = sorted(vocabs)
+        for unit in vocabs_sorted:
+            fout.write(unit + "\n")
+
+        print(f"bpe vocab size: {len(vocabs_sorted)}")
+        print(f"skip {stats['num_empty']} empty lines")
+        print(f"filter {stats['num_filtered']} invalid lines")
+
+    fout.write(SOS + "\n") # <sos/eos>
+    fout.close()
 
 
 if __name__ == '__main__':

utils/format_data.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+"""format manifest with more metadata."""
+import argparse
+import functools
+import json
+from collections import Counter
+import os
+import copy
+import tempfile
+
+from deepspeech.frontend.utility import read_manifest
+from deepspeech.frontend.utility import UNK
+from deepspeech.frontend.utility import BLANK
+from deepspeech.frontend.utility import SOS
+from deepspeech.utils.utility import add_arguments
+from deepspeech.utils.utility import print_arguments
+
+parser = argparse.ArgumentParser(description=__doc__)
+add_arg = functools.partial(add_arguments, argparser=parser)
+# yapf: disable
+add_arg('feat_type', str, "raw", "speech feature type, e.g. raw(wav, flac), kaldi")
+add_arg('unit_type', str, "character", "Unit type, e.g. character, word, bpe")
+add_arg('vocab_path',       str,
+        'examples/librispeech/data/vocab.txt',
+        "Filepath to write the vocabulary.")
+add_arg('manifest_paths',   str,
+        None,
+        "Filepaths of manifests for building vocabulary. "
+        "You can provide multiple manifest files.",
+        nargs='+',
+        required=True)
+# bpe
+add_arg('bpe_model_prefix', str, "bpe_model_%(bpe_mode)_%(count_threshold)", "bpe model prefix, only need when `unit_type` is bpe")
+add_arg('output_path',  str, None, "filepath of formated manifest.", required=True)
+# yapf: disable
+args = parser.parse_args()
+
+
+def main():
+    print_arguments(args)
+
+    # read vocab
+    vocab = dict()
+    with open(args.vocab_path, 'r', encoding='utf-8') as fin:
+        for line in fin:
+            token = line.strip()
+            vocab[token] = len(vocab)
+    vocab_size = len(vocab)
+
+    fout = open(args.output_path, 'w', encoding='utf-8')
+
+    if args.unit_type != 'bpe':
+        for manifest_path in args.manifest_paths:
+            manifest_jsons = read_manifest(manifest_path)
+            for line_json in manifest_jsons:
+                tokens = []
+                tokenids = []
+                if args.unit_type == 'character':
+                    for char in line_json['text']:
+                        tokens.append(char)
+                        tokenids.append(vocab[char])
+                elif args.unit_type == 'word':
+                    for word in line_json['text'].split():
+                        tokens.append(word)
+                        tokenids.append(vocab[word])
+                line_json['token'] = tokens
+                line_json['token_id'] = tokenids
+                line_json['token_shape'] = (len(tokenids), vocab_size)
+                fout.write(json.dumps(line_json) + '\n')
+    else:
+        import sentencepiece as spm
+
+        # encode
+        sp = spm.SentencePieceProcessor()
+        sp.Load(args.bpe_model_prefix + '.model')
+
+        def valid(line):
+            return True
+
+        def encode(l):
+            return sp.EncodeAsPieces(l)
+
+        def encode_line(line):
+            line = line.strip()
+            if len(line) > 0:
+                line = encode(line)
+                if valid(line):
+                    return line
+                else:
+                    stats["num_filtered"] += 1
+            else:
+                stats["num_empty"] += 1
+            return None
+
+        for manifest_path in args.manifest_paths:
+            manifest_jsons = read_manifest(manifest_path)
+            for line_json in manifest_jsons:
+                line = line_json['text']
+                tokens = []
+                tokenids = []
+                enc_line = encode_line(line)
+                for code in enc_line:
+                    tokens.append(code)
+                    tokenids.append(vocab[code])
+                    #print(code, vocab[code])
+                line_json['token'] = tokens
+                line_json['token_id'] = tokenids
+                line_json['token_shape'] = (len(tokenids), vocab_size)
+                fout.write(json.dumps(line_json) + '\n')
+
+    fout.close()
+
+
+if __name__ == '__main__':
+    main()