def _read_audio(
cut: Cut,
suppress_errors: bool = False,
recording_field: Optional[str] = None) -> Optional[torch.Tensor]:
"""
Loads audio data from cut, or returns None if there was an error
and ``suppress_errors`` was set to ``True``.
"""
with suppress_and_warn(
AudioLoadingError,
DurationMismatchError,
NonPositiveEnergyError,
enabled=suppress_errors,
):
if recording_field is None:
audio = cut.load_audio()
else:
attr = getattr(cut, recording_field)
assert isinstance(
attr, Recording
), f"Expected 'getattr(cut, {recording_field})' to yield Recording, got {type(attr)}"
audio = cut.load_custom(recording_field)
assert audio.shape[
0] == 1, f"Expected single-channel audio in cut:\n{cut}"
return torch.from_numpy(audio[0])
def generate_cut(cuts: CutSet):
for cut in cuts:
with suppress_and_warn(AudioLoadingError,
DurationMismatchError):
yield cut.compute_features(
extractor=self.feature_extractor,
augment_fn=self.augment_fn,
)
def _read_audio(cut: Cut,
suppress_errors: bool = False) -> Optional[torch.Tensor]:
"""
Loads audio data from cut, or returns None if there was an error
and ``suppress_errors`` was set to ``True``.
"""
with suppress_and_warn(AudioLoadingError,
DurationMismatchError,
enabled=suppress_errors):
return torch.from_numpy(cut.load_audio()[0])
def write(self, manifest: MonoCut) -> bool:
"""
Converts a Cut to a dict, pickles it, and then stores into a tarfile.
:param manifest: the manifest to be written.
:return: bool indicating whether the writing was successful.
"""
with suppress_and_warn(Exception, enabled=self.fault_tolerant):
cut = manifest.move_to_memory(
audio_format=self.audio_format,
load_audio=self.load_audio,
load_features=self.load_features,
load_custom=self.load_custom,
)
data = pickle.dumps(cut.to_dict())
self.writer.write({"__key__": cut.id, "data": data})
return True
# Will get here if an exception happened.
return False
def __getitem__(self, cuts: CutSet) -> Dict[str, Any]:
if self.collate:
audio, audio_lens = collate_audio(cuts)
return {
"cuts": cuts,
"audio": audio,
"audio_lens": audio_lens,
}
else:
remain_cuts = []
remain_audios = []
for c in cuts:
with suppress_and_warn(AudioLoadingError,
DurationMismatchError):
remain_audios.append(c.load_audio())
remain_cuts.append(c)
return {
"cuts": CutSet.from_cuts(remain_cuts),
"audio": remain_audios
}