def main(
config: str = DEFAULT_YAML,
h5: str = None,
subwords: bool = False,
sentence_piece: bool = False,
output: str = None,
):
assert h5 and output
tf.keras.backend.clear_session()
tf.compat.v1.enable_control_flow_v2()
config = Config(config)
speech_featurizer, text_featurizer = featurizer_helpers.prepare_featurizers(
config=config,
subwords=subwords,
sentence_piece=sentence_piece,
)
deepspeech2 = DeepSpeech2(**config.model_config,
vocabulary_size=text_featurizer.num_classes)
deepspeech2.make(speech_featurizer.shape)
deepspeech2.load_weights(h5, by_name=True)
deepspeech2.summary(line_length=100)
deepspeech2.add_featurizers(speech_featurizer, text_featurizer)
exec_helpers.convert_tflite(model=deepspeech2, output=output)
def main(
config: str = DEFAULT_YAML,
saved: str = None,
mxp: bool = False,
bs: int = None,
sentence_piece: bool = False,
subwords: bool = False,
device: int = 0,
cpu: bool = False,
output: str = "test.tsv",
):
assert saved and output
tf.random.set_seed(0)
tf.keras.backend.clear_session()
tf.config.optimizer.set_experimental_options({"auto_mixed_precision": mxp})
env_util.setup_devices([device], cpu=cpu)
config = Config(config)
speech_featurizer, text_featurizer = featurizer_helpers.prepare_featurizers(
config=config,
subwords=subwords,
sentence_piece=sentence_piece,
)
deepspeech2 = DeepSpeech2(**config.model_config,
vocabulary_size=text_featurizer.num_classes)
deepspeech2.make(speech_featurizer.shape)
deepspeech2.load_weights(saved, by_name=True)
deepspeech2.summary(line_length=100)
deepspeech2.add_featurizers(speech_featurizer, text_featurizer)
test_dataset = dataset_helpers.prepare_testing_datasets(
config=config,
speech_featurizer=speech_featurizer,
text_featurizer=text_featurizer)
batch_size = bs or config.learning_config.running_config.batch_size
test_data_loader = test_dataset.create(batch_size)
exec_helpers.run_testing(model=deepspeech2,
test_dataset=test_dataset,
test_data_loader=test_data_loader,
output=output)
def test_ds2():
config = Config(DEFAULT_YAML)
text_featurizer = CharFeaturizer(config.decoder_config)
speech_featurizer = TFSpeechFeaturizer(config.speech_config)
model = DeepSpeech2(vocabulary_size=text_featurizer.num_classes,
**config.model_config)
model.make(speech_featurizer.shape)
model.summary(line_length=150)
model.add_featurizers(speech_featurizer=speech_featurizer,
text_featurizer=text_featurizer)
concrete_func = model.make_tflite_function(
greedy=False).get_concrete_function()
converter = tf.lite.TFLiteConverter.from_concrete_functions(
[concrete_func])
converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.experimental_new_converter = True
converter.target_spec.supported_ops = [
tf.lite.OpsSet.TFLITE_BUILTINS, tf.lite.OpsSet.SELECT_TF_OPS
]
converter.convert()
print("Converted successfully with beam search")
concrete_func = model.make_tflite_function(
greedy=True).get_concrete_function()
converter = tf.lite.TFLiteConverter.from_concrete_functions(
[concrete_func])
converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.experimental_new_converter = True
converter.target_spec.supported_ops = [
tf.lite.OpsSet.TFLITE_BUILTINS, tf.lite.OpsSet.SELECT_TF_OPS
]
converter.convert()
print("Converted successfully with greedy")
def main(
config: str = DEFAULT_YAML,
tfrecords: bool = False,
sentence_piece: bool = False,
subwords: bool = False,
bs: int = None,
spx: int = 1,
metadata: str = None,
static_length: bool = False,
devices: list = [0],
mxp: bool = False,
pretrained: str = None,
):
tf.keras.backend.clear_session()
tf.config.optimizer.set_experimental_options({"auto_mixed_precision": mxp})
strategy = env_util.setup_strategy(devices)
config = Config(config)
speech_featurizer, text_featurizer = featurizer_helpers.prepare_featurizers(
config=config,
subwords=subwords,
sentence_piece=sentence_piece,
)
train_dataset, eval_dataset = dataset_helpers.prepare_training_datasets(
config=config,
speech_featurizer=speech_featurizer,
text_featurizer=text_featurizer,
tfrecords=tfrecords,
metadata=metadata,
)
if not static_length:
speech_featurizer.reset_length()
text_featurizer.reset_length()
train_data_loader, eval_data_loader, global_batch_size = dataset_helpers.prepare_training_data_loaders(
config=config,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
strategy=strategy,
batch_size=bs,
)
with strategy.scope():
deepspeech2 = DeepSpeech2(**config.model_config,
vocabulary_size=text_featurizer.num_classes)
deepspeech2.make(speech_featurizer.shape, batch_size=global_batch_size)
if pretrained:
deepspeech2.load_weights(pretrained,
by_name=True,
skip_mismatch=True)
deepspeech2.summary(line_length=100)
deepspeech2.compile(
optimizer=config.learning_config.optimizer_config,
experimental_steps_per_execution=spx,
global_batch_size=global_batch_size,
blank=text_featurizer.blank,
)
callbacks = [
tf.keras.callbacks.ModelCheckpoint(
**config.learning_config.running_config.checkpoint),
tf.keras.callbacks.experimental.BackupAndRestore(
config.learning_config.running_config.states_dir),
tf.keras.callbacks.TensorBoard(
**config.learning_config.running_config.tensorboard),
]
deepspeech2.fit(
train_data_loader,
epochs=config.learning_config.running_config.num_epochs,
validation_data=eval_data_loader,
callbacks=callbacks,
steps_per_epoch=train_dataset.total_steps,
validation_steps=eval_dataset.total_steps
if eval_data_loader else None,
)
help="TFLite file path to be exported")
args = parser.parse_args()
assert args.saved and args.output
config = Config(args.config)
speech_featurizer = TFSpeechFeaturizer(config.speech_config)
if args.subwords:
text_featurizer = SubwordFeaturizer(config.decoder_config)
else:
text_featurizer = CharFeaturizer(config.decoder_config)
# build model
deepspeech2 = DeepSpeech2(**config.model_config,
vocabulary_size=text_featurizer.num_classes)
deepspeech2.make(speech_featurizer.shape)
deepspeech2.load_weights(args.saved)
deepspeech2.summary(line_length=100)
deepspeech2.add_featurizers(speech_featurizer, text_featurizer)
concrete_func = deepspeech2.make_tflite_function().get_concrete_function()
converter = tf.lite.TFLiteConverter.from_concrete_functions([concrete_func])
converter.experimental_new_converter = True
converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.target_spec.supported_ops = [
tf.lite.OpsSet.TFLITE_BUILTINS, tf.lite.OpsSet.SELECT_TF_OPS
]
tflite_model = converter.convert()
args.output = file_util.preprocess_paths(args.output)