admin-valentin/WhisperLiveKit
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Enhanced use_punctuation_split for diarization. further improvements still needed

Browse source
This commit is contained in:
Quentin Fuxa 2025-07-16 12:06:17 +02:00
parent 12a544164f
commit 5b9977c9af
1 changed files with 18 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

30
whisperlivekit/diarization/diarization_online.py
View file

@ -165,7 +165,7 @@ class WebSocketAudioSource(AudioSource):
class DiartDiarization: class DiartDiarization:
def __init__(self, sample_rate: int = 16000, config : SpeakerDiarizationConfig = None, use_microphone: bool = False, block_duration: float = 0.5, segmentation_model_name: str = "pyannote/segmentation-3.0", embedding_model_name: str = "speechbrain/spkrec-ecapa-voxceleb"): def __init__(self, sample_rate: int = 16000, config : SpeakerDiarizationConfig = None, use_microphone: bool = False, block_duration: float = 1.5, segmentation_model_name: str = "pyannote/segmentation-3.0", embedding_model_name: str = "pyannote/embedding"):
segmentation_model = m.SegmentationModel.from_pretrained(segmentation_model_name) segmentation_model = m.SegmentationModel.from_pretrained(segmentation_model_name)
embedding_model = m.EmbeddingModel.from_pretrained(embedding_model_name) embedding_model = m.EmbeddingModel.from_pretrained(embedding_model_name)
@ -206,8 +206,7 @@ class DiartDiarization:
""" """
if self.custom_source: if self.custom_source:
self.custom_source.push_audio(pcm_array) self.custom_source.push_audio(pcm_array)
self.observer.clear_old_segments() # self.observer.clear_old_segments()
return self.observer.get_segments()
def close(self): def close(self):
"""Close the audio source.""" """Close the audio source."""
@ -231,7 +230,15 @@ class DiartDiarization:
if not self.lag_diart and segments and tokens: if not self.lag_diart and segments and tokens:
self.lag_diart = segments[0].start - tokens[0].start self.lag_diart = segments[0].start - tokens[0].start
tokens = add_speaker_to_tokens(segments, tokens)
if not use_punctuation_split:
for token in tokens:
for segment in segments:
if not (segment.end <= token.start + self.lag_diart or segment.start >= token.end + self.lag_diart):
token.speaker = extract_number(segment.speaker) + 1
end_attributed_speaker = max(token.end, end_attributed_speaker)
else:
tokens = add_speaker_to_tokens(segments, tokens)
return tokens return tokens
def concatenate_speakers(segments): def concatenate_speakers(segments):
@ -242,16 +249,15 @@ def concatenate_speakers(segments):
segments_concatenated.append({"speaker": speaker, "begin": segment.start, "end": segment.end}) segments_concatenated.append({"speaker": speaker, "begin": segment.start, "end": segment.end})
else: else:
segments_concatenated[-1]['end'] = segment.end segments_concatenated[-1]['end'] = segment.end
print("Segments concatenated:") # print("Segments concatenated:")
for entry in segments_concatenated: # for entry in segments_concatenated:
print(f"Speaker {entry['speaker']}: {entry['begin']:.2f}s - {entry['end']:.2f}s") # print(f"Speaker {entry['speaker']}: {entry['begin']:.2f}s - {entry['end']:.2f}s")
return segments_concatenated return segments_concatenated
def add_speaker_to_tokens(segments, tokens): def add_speaker_to_tokens(segments, tokens):
""" """
Assign speakers to tokens based on diarization segments, with punctuation-aware boundary adjustment. Assign speakers to tokens based on diarization segments, with punctuation-aware boundary adjustment.
Refactored for clarity; behavior unchanged.
""" """
punctuation_marks = {'.', '!', '?'} punctuation_marks = {'.', '!', '?'}
punctuation_tokens = [token for token in tokens if token.text.strip() in punctuation_marks] punctuation_tokens = [token for token in tokens if token.text.strip() in punctuation_marks]
@ -284,10 +290,10 @@ def add_speaker_to_tokens(segments, tokens):
if token.end <= segment['end']: if token.end <= segment['end']:
token.speaker = segment['speaker'] token.speaker = segment['speaker']
ind_last_speaker = i + 1 ind_last_speaker = i + 1
print( # print(
f"Token '{token.text}' ('begin': {token.start:.2f}, 'end': {token.end:.2f}) " # f"Token '{token.text}' ('begin': {token.start:.2f}, 'end': {token.end:.2f}) "
f"assigned to Speaker {segment['speaker']} ('segment': {segment['begin']:.2f}-{segment['end']:.2f})" # f"assigned to Speaker {segment['speaker']} ('segment': {segment['begin']:.2f}-{segment['end']:.2f})"
) # )
elif token.start > segment['end']: elif token.start > segment['end']:
break break
return tokens return tokens