Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

add 'decode_options' to annotate_with_whisper #954

Merged
merged 4 commits into from
Jan 24, 2023
Merged

add 'decode_options' to annotate_with_whisper #954

Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
c757cc5
add decode_options argument to whisper
Jan 23, 2023
ff7f4db
Merge branch 'lhotse-speech:master' into master
Joemgu7 Jan 23, 2023
53c3e85
add download root for whisper in lhotse
Jan 24, 2023
1d98fb2
add options to docstring
Jan 24, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 12 additions & 9 deletions lhotse/workflows/whisper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,10 +17,11 @@

def annotate_with_whisper(
manifest: Union[RecordingSet, CutSet],
language: Optional[str] = None,
model_name: str = "base",
device: str = "cpu",
force_nonoverlapping: bool = False,
download_root: Optional[str] = None,
Joemgu7 marked this conversation as resolved.
Show resolved Hide resolved
**decode_options,
) -> Generator[MonoCut, None, None]:
"""
Use OpenAI Whisper model to annotate either RECORDINGS_MANIFEST, RECORDINGS_DIR, or CUTS_MANIFEST.
Expand Down Expand Up @@ -48,29 +49,30 @@ def annotate_with_whisper(

if isinstance(manifest, RecordingSet):
yield from _annotate_recordings(
manifest, language, model_name, device, force_nonoverlapping
manifest, model_name, device, force_nonoverlapping, download_root, **decode_options
)
elif isinstance(manifest, CutSet):
yield from _annotate_cuts(
manifest, language, model_name, device, force_nonoverlapping
manifest, model_name, device, force_nonoverlapping, download_root, **decode_options
)
else:
raise ValueError("The ``manifest`` must be either a RecordingSet or a CutSet.")


def _annotate_recordings(
recordings: RecordingSet,
language: str,
model_name: str,
device: str,
force_nonoverlapping: bool,
download_root: Optional[str] = None,
**decode_options,
):
"""
Helper function that annotates a RecordingSet with Whisper.
"""
import whisper

model = whisper.load_model(model_name, device=device)
model = whisper.load_model(model_name, device=device, download_root=download_root)

for recording in recordings:
if recording.num_channels > 1:
Expand All @@ -80,7 +82,7 @@ def _annotate_recordings(
)
continue
audio = torch.from_numpy(recording.resample(16000).load_audio()).squeeze(0)
result = whisper.transcribe(model=model, audio=audio, language=language)
result = whisper.transcribe(model=model, audio=audio, **decode_options)
# Create supervisions from segments while filtering out those with negative duration.
supervisions = [
SupervisionSegment(
Expand Down Expand Up @@ -113,17 +115,18 @@ def _annotate_recordings(

def _annotate_cuts(
cuts: CutSet,
language: str,
model_name: str,
device: str,
force_nonoverlapping: bool,
download_root: Optional[str] = None,
**decode_options,
):
"""
Helper function that annotates a CutSet with Whisper.
"""
import whisper

model = whisper.load_model(model_name, device=device)
model = whisper.load_model(model_name, device=device, download_root=download_root)

for cut in cuts:
if cut.num_channels > 1:
Expand All @@ -133,7 +136,7 @@ def _annotate_cuts(
)
continue
audio = torch.from_numpy(cut.resample(16000).load_audio()).squeeze(0)
result = whisper.transcribe(model=model, audio=audio, language=language)
result = whisper.transcribe(model=model, audio=audio, **decode_options)
# Create supervisions from segments while filtering out those with negative duration.
supervisions = [
SupervisionSegment(
Expand Down