)
estimator_spec = tf.estimator.EstimatorSpec(
mode = mode, loss = loss, train_op = train_op
)
elif mode == tf.estimator.ModeKeys.EVAL:
estimator_spec = tf.estimator.EstimatorSpec(
mode = tf.estimator.ModeKeys.EVAL, loss = loss
)
return estimator_spec
train_hooks = [tf.train.LoggingTensorHook(['total_loss'], every_n_iter = 1)]
train_dataset = get_dataset()
save_directory = 'speaker-split-fast-swave'
train.run_training(
train_fn = train_dataset,
model_fn = model_fn,
model_dir = save_directory,
num_gpus = 1,
log_step = 1,
save_checkpoint_step = 3000,
max_steps = total_steps,
train_hooks = train_hooks,
eval_step = 0,
)
return estimator_spec
train_hooks = [
tf.train.LoggingTensorHook(
[
'loss',
'duration_loss',
'mel_loss_before',
'mel_loss_after',
'energies_loss',
'f0s_loss',
],
every_n_iter=1,
)
]
train_dataset = get_dataset(files)
train.run_training(
train_fn=train_dataset,
model_fn=model_fn,
model_dir='fastspeech2-husein-v2',
num_gpus=1,
log_step=1,
save_checkpoint_step=2000,
max_steps=total_steps,
eval_fn=None,
train_hooks=train_hooks,
)
elif mode == tf.estimator.ModeKeys.EVAL:
estimator_spec = tf.estimator.EstimatorSpec(
mode=tf.estimator.ModeKeys.EVAL,
loss=loss,
eval_metric_ops={'accuracy': accuracy},
)
return estimator_spec
train_hooks = [
tf.train.LoggingTensorHook(['train_accuracy', 'train_loss'],
every_n_iter=1)
]
train_dataset = get_dataset()
save_directory = 'output-speakernet-speaker-count'
train.run_training(
train_fn=train_dataset,
model_fn=model_fn,
model_dir=save_directory,
num_gpus=1,
log_step=1,
save_checkpoint_step=25000,
max_steps=300_000,
train_hooks=train_hooks,
)
loss = loss,
eval_metric_ops = {
'accuracy': ctc.metrics.ctc_sequence_accuracy_estimator(
logits, targets_int32, seq_lens
)
},
)
return estimator_spec
train_hooks = [
tf.train.LoggingTensorHook(
['train_accuracy', 'train_loss'], every_n_iter = 1
)
]
train_dataset = get_dataset('training-librispeech/data/librispeech-train-*')
dev_dataset = get_dataset('training-librispeech/data/librispeech-dev-*')
train.run_training(
train_fn = train_dataset,
model_fn = model_fn,
model_dir = 'asr-quartznet-librispeech-adam',
num_gpus = 2,
log_step = 1,
save_checkpoint_step = parameters['lr_policy_params']['warmup_steps'],
max_steps = parameters['lr_policy_params']['decay_steps'],
eval_fn = dev_dataset,
train_hooks = train_hooks,
)
return estimator_spec
train_hooks = [
tf.train.LoggingTensorHook(
[
'loss',
'stop_token_loss',
'mel_loss_before',
'mel_loss_after',
'loss_att',
],
every_n_iter = 1,
)
]
train_dataset = get_dataset(files['train'])
dev_dataset = get_dataset(files['test'])
train.run_training(
train_fn = train_dataset,
model_fn = model_fn,
model_dir = 'tacotron2-female-3',
num_gpus = 1,
log_step = 1,
save_checkpoint_step = 5000,
max_steps = num_train_steps,
eval_fn = dev_dataset,
train_hooks = train_hooks,
)
)
elif mode == tf.estimator.ModeKeys.EVAL:
estimator_spec = tf.estimator.EstimatorSpec(
mode = tf.estimator.ModeKeys.EVAL, loss = loss
)
return estimator_spec
train_hooks = [tf.train.LoggingTensorHook(['train_loss'], every_n_iter = 1)]
train_dataset = get_dataset(
'../speech-bahasa/bahasa-asr/data/bahasa-asr-train-*'
)
dev_dataset = get_dataset(
'../speech-bahasa/bahasa-asr-test/data/bahasa-asr-dev-*'
)
train.run_training(
train_fn = train_dataset,
model_fn = model_fn,
model_dir = 'asr-base-conformer-transducer',
num_gpus = 2,
log_step = 1,
save_checkpoint_step = 5000,
max_steps = 500_000,
eval_fn = dev_dataset,
train_hooks = train_hooks,
)
return estimator_spec
train_hooks = [
tf.train.LoggingTensorHook(
[
'loss',
'stop_token_loss',
'mel_loss_before',
'mel_loss_after',
'loss_att',
],
every_n_iter = 1,
)
]
train_dataset = get_dataset(files)
train.run_training(
train_fn = train_dataset,
model_fn = model_fn,
model_dir = 'tacotron2-case-haqkiem',
num_gpus = 1,
log_step = 1,
save_checkpoint_step = 2000,
max_steps = 100000,
eval_fn = None,
train_hooks = train_hooks,
)
eval_metric_ops={
'accuracy':
ctc.metrics.ctc_sequence_accuracy_estimator(
logits, targets_int32, seq_lens)
},
)
return estimator_spec
train_hooks = [
tf.train.LoggingTensorHook(['train_accuracy', 'train_loss'],
every_n_iter=1)
]
train_dataset = get_dataset(
'../speech-bahasa/bahasa-asr/data/bahasa-asr-train-*')
dev_dataset = get_dataset(
'../speech-bahasa/bahasa-asr-test/data/bahasa-asr-dev-*')
train.run_training(
train_fn=train_dataset,
model_fn=model_fn,
model_dir='asr-mini-jasper-ctc',
num_gpus=2,
log_step=1,
save_checkpoint_step=5000,
max_steps=parameters['lr_policy_params']['decay_steps'],
eval_fn=dev_dataset,
train_hooks=train_hooks,
)
)
return estimator_spec
train_hooks = [
tf.train.LoggingTensorHook(
['train_accuracy', 'train_loss'], every_n_iter = 1
)
]
train_files = glob('vad2/data/vad-train-*') + glob('noise/data/vad-train-*')
train_dataset = get_dataset(train_files, batch_size = 32)
dev_files = glob('vad2/data/vad-dev-*') + glob('noise/data/vad-dev-*')
dev_dataset = get_dataset(dev_files, batch_size = 16)
save_directory = 'output-inception-v4-vad'
train.run_training(
train_fn = train_dataset,
model_fn = model_fn,
model_dir = save_directory,
num_gpus = 1,
log_step = 1,
save_checkpoint_step = 25000,
max_steps = epochs,
eval_fn = dev_dataset,
train_hooks = train_hooks,
)
