Merge pull request #398 from AlexStansfield/feature/faster-whisper-1.2.0

feat: update to support faster whisper 1.2.0

whisper_live/backend/faster_whisper_backend.py

@@ -76,7 +76,7 @@ class ServeClientFasterWhisper(ServeClientBase):
         self.language = "en" if self.model_size_or_path.endswith("en") else language
         self.task = task
         self.initial_prompt = initial_prompt
-        self.vad_parameters = vad_parameters or {"onset": 0.5}
+        self.vad_parameters = vad_parameters or {"threshold": 0.5}
 
         device = "cuda" if torch.cuda.is_available() else "cpu"
         if device == "cuda":

whisper_live/transcriber/transcriber_faster_whisper.py

@@ -27,7 +27,6 @@ from faster_whisper.vad import (
     VadOptions,
     collect_chunks,
     get_speech_timestamps,
-    merge_segments,
 )
 
 
@@ -407,8 +406,7 @@ class BatchedInferencePipeline:
                         **vad_parameters, max_speech_duration_s=chunk_length
                     )
 
-                active_segments = get_speech_timestamps(audio, vad_parameters)
-                clip_timestamps = merge_segments(active_segments, vad_parameters)
+                clip_timestamps = get_speech_timestamps(audio, vad_parameters)
             # run the audio if it is less than 30 sec even without clip_timestamps
             elif duration < chunk_length:
                 clip_timestamps = [{"start": 0, "end": audio.shape[0]}]