update text threhold

transcribe/whisper_llm_serve.py

@@ -55,6 +55,7 @@ class WhisperTranscriptionService:
         self.frames_np = None
         #  完整音频队列
         self.segments_queue = collections.deque()
+        self._temp_string = ""
 
         self._transcrible_analysis = None
         # 启动处理线程
@@ -136,14 +137,14 @@ class WhisperTranscriptionService:
                         self.frames_np = frame_np.copy()
                     else:
                         self.frames_np = np.append(self.frames_np, frame_np)
-                    if speech_status == "END" and len(self.frames_np) > 0:
+                    if speech_status == "END" and len(frame_np) > 0:
                         self.segments_queue.appendleft(self.frames_np.copy())
                         self.frames_np = np.array([], dtype=np.float32)
             except queue.Empty:
                 pass
 
-    def _process_transcription_results_2(self, segments: List[TranscriptToken],partial):
-        seg_text = self.text_separator.join(seg.text for seg in segments)
+    def _process_transcription_results_2(self, seg_text:str,partial):
+    
         item =  TransResult(
                 seg_id=self.row_number,
                 context=seg_text,
@@ -181,6 +182,7 @@ class WhisperTranscriptionService:
                 silence_audio = np.zeros(self.sample_rate, dtype=np.float32)
                 silence_audio[-len(audio_buffer):] = audio_buffer
                 audio_buffer = silence_audio
+        
 
             logger.debug(f"audio buffer size: {len(audio_buffer) / self.sample_rate:.2f}s")
             # try:
@@ -188,14 +190,26 @@ class WhisperTranscriptionService:
             segments = meta_item.segments
             logger.debug(f"Segments: {segments}")
             if len(segments):
-                result = self._process_transcription_results_2(segments, partial)
+                seg_text = self.text_separator.join(seg.text for seg in segments)
+                if self._temp_string:
+                    seg_text = self._temp_string + seg_text
+
+                if partial == False:
+                    if len(seg_text) < config.TEXT_THREHOLD:
+                        partial = True
+                        self._temp_string = seg_text
+                    else:
+                        self._temp_string = ""
+                    
+            
+                result = self._process_transcription_results_2(seg_text, partial)
                 self._send_result_to_client(result)
                 time.sleep(0.1)
 
-            if partial == False:
-                frame_epoch = 1
-            else:
-                frame_epoch += 1
+                if partial == False:
+                    frame_epoch = 1
+                else:
+                    frame_epoch += 1
             # 处理转录结果并发送到客户端
             # for result in self._process_transcription_results(segments, audio_buffer):
             #     self._send_result_to_client(result)