def test_featurizer():
config = {
"output_path_prefix":
"/data/models/asr/conformer_sentencepiece_subword",
"model_type":
"unigram",
"target_vocab_size":
8000,
"blank_at_zero":
True,
"beam_width":
5,
"norm_score":
True,
"corpus_files": [
"/data/datasets/LibriSpeech/train-clean-100/transcripts.tsv"
"/data/datasets/LibriSpeech/train-clean-360/transcripts.tsv"
"/data/datasets/LibriSpeech/train-other-500/transcripts.tsv"
]
}
config_speech = {
"sample_rate": 16000,
"frame_ms": 25,
"stride_ms": 10,
"num_feature_bins": 80,
'feature_type': "log_mel_spectrogram",
"preemphasis": 0.97,
"normalize_signal": True,
"normalize_feature": True,
"normalize_per_feature": False
}
text_featurizer_sentencepiece = SentencePieceFeaturizer.load_from_file(
config, None)
subwords_path = os.path.join(os.path.abspath(os.path.dirname(__file__)),
os.pardir, os.pardir, "vocabularies",
"librispeech_train_4_1030.subwords")
text_featurizer_subwords = SubwordFeaturizer.load_from_file(
config, subwords_path)
speech_featurizer = TFSpeechFeaturizer(config_speech)
data_path = os.path.join(os.path.abspath(os.path.dirname(__file__)),
"transcripts_librispeech_train_clean_100.tsv")
def get_data(featurizer: TextFeaturizer):
train_dataset = ASRSliceDataset(data_paths=[data_path],
speech_featurizer=speech_featurizer,
text_featurizer=featurizer,
stage="train",
shuffle=False)
train_data = train_dataset.create(1)
return next(iter(train_data))
data_sentencepiece = get_data(text_featurizer_sentencepiece)
data_subwords = get_data(text_featurizer_subwords)
assert len(data_sentencepiece) == len(data_subwords)
assert data_sentencepiece[0].shape == data_subwords[0].shape
assert data_sentencepiece[0].dtype == data_subwords[0].dtype
def test_iextract():
config = {
"output_path_prefix": "/data/models/asr/conformer_sentencepiece_subword",
"model_type": "unigram",
"target_vocab_size": 8000,
"blank_at_zero": True,
"beam_width": 5,
"norm_score": True,
"corpus_files": [
"/data/datasets/LibriSpeech/train-clean-100/transcripts.tsv"
"/data/datasets/LibriSpeech/train-clean-360/transcripts.tsv"
"/data/datasets/LibriSpeech/train-other-500/transcripts.tsv"]}
config_speech = {
"sample_rate": 16000,
"frame_ms": 25,
"stride_ms": 10,
"num_feature_bins": 80,
'feature_type': "log_mel_spectrogram",
"preemphasis": 0.97,
"normalize_signal": True,
"normalize_feature": True,
"normalize_per_frame": False}
text_featurizer_sentencepiece = SentencePieceFeaturizer.load_from_file(config, None)
speech_featurizer = TFSpeechFeaturizer(config_speech)
data_path = os.path.join(os.path.abspath(os.path.dirname(__file__)), "transcripts_librispeech_train_clean_100.tsv")
train_dataset = ASRSliceTestDataset(
data_paths=[data_path],
speech_featurizer=speech_featurizer,
text_featurizer=text_featurizer_sentencepiece,
stage="train",
shuffle=False
)
train_data = train_dataset.create(1)
batch = next(iter(train_data))
file_paths, features, input_length, labels = batch
labels = text_featurizer_sentencepiece.iextract(labels)
labels = labels.numpy()[0].decode("utf-8")
# Open transcript
file_path = file_paths[0].numpy().decode("utf-8")
file_path = re.sub(r"(?<!\s)-[0-9]{4}.flac", ".trans.txt", file_path)
print(file_path)
with open(file_path, "r") as f:
lines = f.read().splitlines()
m = re.search(r"[0-9]+-[0-9]+-[0-9]+\s+([\w\s]+)", lines[0])
transcript = m.groups(1)[0].lower()
assert(labels == transcript)
parser.add_argument("transcripts",
nargs="+",
type=str,
default=None,
help="Paths to transcript files")
args = parser.parse_args()
transcripts = preprocess_paths(args.transcripts)
tfrecords_dir = preprocess_paths(args.tfrecords_dir)
config = Config(args.config)
if args.sentence_piece:
print("Loading SentencePiece model ...")
text_featurizer = SentencePieceFeaturizer.load_from_file(
config.decoder_config, args.subwords)
elif args.subwords and os.path.exists(args.subwords):
print("Loading subwords ...")
text_featurizer = SubwordFeaturizer.load_from_file(config.decoder_config,
args.subwords)
ASRTFRecordDataset(data_paths=transcripts,
tfrecords_dir=tfrecords_dir,
speech_featurizer=None,
text_featurizer=text_featurizer,
stage=args.mode,
shuffle=args.shuffle,
tfrecords_shards=args.tfrecords_shards).create_tfrecords()
env_util.setup_devices([args.device], cpu=args.cpu)
from tensorflow_asr.configs.config import Config
from tensorflow_asr.datasets.asr_dataset import ASRTFRecordDataset, ASRSliceDataset
from tensorflow_asr.featurizers.speech_featurizers import TFSpeechFeaturizer
from tensorflow_asr.featurizers.text_featurizers import SubwordFeaturizer, SentencePieceFeaturizer, CharFeaturizer
from tensorflow_asr.models.ctc.jasper import Jasper
from tensorflow_asr.utils import app_util
config = Config(args.config)
speech_featurizer = TFSpeechFeaturizer(config.speech_config)
if args.sentence_piece:
print("Use SentencePiece ...")
text_featurizer = SentencePieceFeaturizer(config.decoder_config)
elif args.subwords:
print("Use subwords ...")
text_featurizer = SubwordFeaturizer(config.decoder_config)
else:
print("Use characters ...")
text_featurizer = CharFeaturizer(config.decoder_config)
tf.random.set_seed(0)
if args.tfrecords:
test_dataset = ASRTFRecordDataset(
speech_featurizer=speech_featurizer,
text_featurizer=text_featurizer,
**vars(config.learning_config.test_dataset_config))
else:
DEFAULT_YAML = os.path.join(os.path.abspath(os.path.dirname(__file__)),
"config.yml")
tf.keras.backend.clear_session()
parser = argparse.ArgumentParser(prog="Vocab Training with SentencePiece")
parser.add_argument("--config",
type=str,
default=DEFAULT_YAML,
help="The file path of model configuration file")
parser.add_argument("--devices",
type=int,
nargs="*",
default=[0],
help="Devices' ids to apply distributed training")
args = parser.parse_args()
strategy = setup_strategy(args.devices)
from tensorflow_asr.configs.config import Config
from tensorflow_asr.featurizers.text_featurizers import SentencePieceFeaturizer
config = Config(args.config)
logger.info("Generating subwords ...")
text_featurizer = SentencePieceFeaturizer.build_from_corpus(
config.decoder_config)
