def override_config():
config = locator.get_config()
config.build = 1000
config.display_interval = 1
config.checkpoint_interval = 2
config.cache_size = config.batch_size * 2
config.verbose = False
def get_trainer(self, stage: int):
return Trainer(locator.get_config(),
locator.get_filesystem_provider(),
locator.get_acoustic_model(),
locator.get_evaluator(),
locator.get_frontend_processor(),
stage=stage)
core_config.__dict__.update(config.__dict__)
# original common config module extends
config.__dict__.update(core_config.__dict__)
return config
def with_generated_params(config):
import vui.frontend.dsp as dsp
config.seg_length = dsp.get_best_segment_length(config.framerate)
config.freq_count = dsp.get_freq_count(config.seg_length)
config.freq_res = dsp.get_freq_resolution(
config.framerate, config.freq_count)
config.frontend_shape = (config.spectrums_per_sec, config.harmonics_count)
return config
config = locator.get_config()
config = with_core_params(config)
config = with_generated_params(config)
if not config.verbose:
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
print("Build {}".format(config.build))
def run(word_handler: Callable, duration: Optional[float] = None):
vui = locator.get_voice_user_interface(word_handler)
vui.run(duration)
def get_wav_folder_storage(dataset_name: str) -> InMemoryStorage:
path = locator.get_filesystem_provider().get_dataset_path(
"r", dataset_name)
max_length = locator.get_config().framerate
return get_storage_from_wav_folder(path, max_length)
def run_training():
with HiddenPrints():
import train
stages = locator.get_config().stages
for stage in range(stages):
train.main(stage, 2)