def __init__(
self,
config,
training=True,
enable_tflite_convertible=False,
speech_config=dict,
):
config['LAS_decoder'].update({'encoder_dim': config['dmodel']})
decoder_config = LASConfig(**config['LAS_decoder'])
super(ConformerLAS, self).__init__(
encoder=ConformerEncoder(
dmodel=config['dmodel'],
reduction_factor=config['reduction_factor'],
num_blocks=config['num_blocks'],
head_size=config['head_size'],
num_heads=config['num_heads'],
kernel_size=config['kernel_size'],
fc_factor=config['fc_factor'],
dropout=config['dropout'],
name=config['name']),
config=decoder_config,
training=training,
enable_tflite_convertible=enable_tflite_convertible,
speech_config=speech_config)
self.time_reduction_factor = config['reduction_factor']
def __init__(self,
config,
training,
name: str = "LAS",
enable_tflite_convertible=False):
config['LAS_decoder'].update({'encoder_dim': config['model_config']['filter_size']})
decoder_config = LASConfig(**config['LAS_decoder'])
super(ESPNetLAS, self).__init__(
encoder= ESPNet(**config['model_config']),
config=decoder_config, training=training,enable_tflite_convertible=enable_tflite_convertible,
name=name)
self.time_reduction_factor = 4
def __init__(self,
config,
input_shape: list,
training,
name: str = "LAS",
enable_tflite_convertible=False):
config['LAS_decoder'].update(
{'encoder_dim': config['fc_conf']['fc_units'][-1]})
decoder_config = LASConfig(**config['LAS_decoder'])
super(DeepSpeech2LAS, self).__init__(
encoder=create_ds2(input_shape=input_shape,
arch_config=config,
name=name),
config=decoder_config,
training=training,
enable_tflite_convertible=enable_tflite_convertible,
)
self.time_reduction_factor = 1
for s in config["conv_conf"]["conv_strides"]:
self.time_reduction_factor *= s[0]
def __init__(self,
config,
input_shape: list,
training,
name: str = "LAS",
enable_tflite_convertible=False,
speech_config=dict):
config['LAS_decoder'].update({'encoder_dim': config['fc_conf']['fc_units'][-1]})
decoder_config = LASConfig(**config['LAS_decoder'])
super(DeepSpeech2LAS, self).__init__(
encoder=DeepSpeech2(arch_config=config,
name=name,
add_wav_info=speech_config['add_wav_info'],
hop_size=int(speech_config['stride_ms'] * speech_config['sample_rate'] // 1000)
),
config=decoder_config, training=training,enable_tflite_convertible=enable_tflite_convertible,
speech_config=speech_config
)
self.time_reduction_factor = 1
for s in config["conv_conf"]["conv_strides"]:
self.time_reduction_factor *= s[0]