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)
def main():
logs_path = locator.get_filesystem_provider().get_logs_dir()
try:
proc = subprocess.Popen(["tensorboard", "--logdir", logs_path])
webbrowser.open("http://localhost:6006/", new=0, autoraise=True)
proc.communicate()
except KeyboardInterrupt:
pass
def main():
status("Starting...")
status("Arranging test experiment...")
experiment_dir = locator.get_filesystem_provider().get_experiment_dir()
with locator.create_isolated_scope():
override_config()
clone_experiment_core(experiment_dir)
status("Running training...")
run_training()
status("Running inference...")
run_inference()
status("Reloading...")
with locator.create_isolated_scope():
override_config()
status("Continuing training...")
run_training()
status("All right!")
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 get_hdf_storage(type_letter: str, dataset_name: str) -> HdfStorage:
path = locator.get_filesystem_provider().get_dataset_path(
type_letter, dataset_name, ".hdf5")
return HdfStorage(path)
import matplotlib.pyplot as plt
import numpy as np
import vui.infrastructure.locator as locator
import vui.dataset.pipeline as pipeline
import vui.frontend.dsp as dsp
config = locator.get_config()
filesystem = locator.get_filesystem_provider()
def input_local():
frames = dsp.read("input.wav")
config.framerate = dsp.get_framerate("input.wav")
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)
return frames
def input() -> np.ndarray:
storage = pipeline.get_hdf_storage('r', "s_en_SpeechCommands")
return np.reshape(storage.fetch_subset("tree", 0, 1), [-1])
def input_bulk(n: int) -> np.ndarray:
storage = pipeline.get_hdf_storage('r', "t_mx_Mix")
return storage.fetch_subset("", 0, n)
def output(frames_before: np.ndarray, frames_after: np.ndarray):
def clone_experiment_core(source_dir: str):
filesystem = locator.get_filesystem_provider()
filesystem.clear_experiment()
filesystem.clone_core_modules(source_dir)