change logic of persist words

transcribe/strategy.py

@@ -250,7 +250,7 @@ class TranscriptStabilityAnalyzer:
             return
 
         # yield from self._handle_short_buffer(current, prev)
-        if buffer_duration <= 12:
+        if buffer_duration < 4:
             yield from self._handle_short_buffer(current, prev)
         else:
             yield from self._handle_long_buffer(current)
@@ -281,8 +281,8 @@ class TranscriptStabilityAnalyzer:
 
     def _handle_long_buffer(self, curr: TranscriptChunk) -> Iterator[TranscriptResult]:
         chunks = curr.split_by(SplitMode.PUNCTUATION)
-        if len(chunks) > 2:
-            stable, remaining = chunks[:-2], chunks[-2:]
+        if len(chunks) > 1:
+            stable, remaining = chunks[:-1], chunks[-1:]
             # stable_str = self.merge_chunks(stable)
             # remaining_str = self.merge_chunks(remaining)
             yield from self._yield_commit_results(

transcribe/whisper_llm_serve.py

@@ -141,8 +141,9 @@ class WhisperTranscriptionService(ServeClientBase):
 
         result = self._translate_pipe.transcrible(audio_buffer.tobytes(), self.source_language)
         segments = result.segments
-
-        log_block("Whisper transcription output", f"{''.join(seg.text for seg in segments)}", "")
+        logger.debug(f"📝 Transcrible: {self.text_separator.join(seg.text for seg in segments)} ")
+        
+        log_block("Whisper transcription output", f"{self.text_separator.join(seg.text for seg in segments)}", "")
         log_block("Whisper transcription time", f"{(time.perf_counter() - start_time):.3f}", "s")
 
         return [
@@ -203,7 +204,7 @@ class WhisperTranscriptionService(ServeClientBase):
             if audio_buffer is None:
                 time.sleep(0.2)
                 continue
-
+            logger.debug(f"🥤 Buffer Length: {len(audio_buffer)/self.sample_rate:.2f} ")
             # c+= 1
             # save_to_wave(f"dev-{c}.wav", audio_buffer)