simulstreaming: fixes token handling during warm-up phase

whisperlivekit/simul_whisper/simul_whisper.py

@@ -261,9 +261,10 @@ class PaddedAlignAttWhisper:
             segments_len -= removed_len
             self.last_attend_frame -= int(TOKENS_PER_SECOND*removed_len)
             self.segments = self.segments[1:]
-            logger.debug(f"remove segments: {len(self.segments)} {len(self.tokens)}")
-            self.context.append_token_ids(self.tokens[1][0,:])
-            self.tokens = [self.initial_tokens] + self.tokens[2:]
+            if len(self.tokens) > 1: # When warming up, we can have a too long segments_len while not having any tokens yet
+                self.context.append_token_ids(self.tokens[1][0,:])
+                self.context.append_token_ids(self.tokens[1][0,:])
+                self.tokens = [self.initial_tokens] + self.tokens[2:]
         return removed_len
 
 

whisperlivekit/whisper_streaming_custom/backends.py

@@ -481,7 +481,6 @@ class SimulStreamingASR(ASRBase):
         try:
             if isinstance(audio, np.ndarray):
                 audio = torch.from_numpy(audio).float()
-            print(audio)
             self.model.insert_audio(audio)
             self.model.infer(True)
             self.model.refresh_segment(complete=True)

whisperlivekit/whisper_streaming_custom/online_asr.py

@@ -680,8 +680,7 @@ class SimulStreamingOnlineProcessor:
 
             
         except Exception as e:
-            logger.error(f"SimulStreaming processing error: {e}")
-            logger.error(f"Error details: {type(e).__name__}: {str(e)}")
+            logger.exception(f"SimulStreaming processing error: {e}")
             return [], self.end
 
     def finish(self) -> Tuple[List[ASRToken], float]: