Merge remote-tracking branch 'upstream/main'

whisper_live/server.py

@@ -408,16 +408,17 @@ class ServeClientTensorRT(ServeClientBase):
             json.dumps(
                 {
                     "uid": self.client_uid,
-                    "message": self.SERVER_READY
+                    "message": self.SERVER_READY,
+                    "backend": "tensorrt"
                 }
             )
         )
 
     def warmup(self, warmup_steps=10):
         logging.info("[INFO:] Warming up TensorRT engine..")
-        mel, duration = self.transcriber.log_mel_spectrogram("tests/jfk.flac")
+        mel, _ = self.transcriber.log_mel_spectrogram("tests/jfk.flac")
         for i in range(warmup_steps):
-            last_segment = self.transcriber.transcribe(mel)
+            self.transcriber.transcribe(mel)
     
     def set_eos(self, eos):
         self.lock.acquire()
@@ -561,7 +562,7 @@ class ServeClientFasterWhisper(ServeClientBase):
         multilingual=False,
         language=None,
         client_uid=None,
-        model="small",
+        model="small.en",
         initial_prompt=None,
         vad_parameters=None,
         ):
@@ -595,6 +596,7 @@ class ServeClientFasterWhisper(ServeClientBase):
         self.task = task
         self.initial_prompt = initial_prompt
         self.vad_parameters = vad_parameters or {"threshold": 0.5}
+        self.no_speech_thresh = 0.45
         
         device = "cuda" if torch.cuda.is_available() else "cpu"
         
@@ -615,7 +617,8 @@ class ServeClientFasterWhisper(ServeClientBase):
             json.dumps(
                 {
                     "uid": self.client_uid,
-                    "message": self.SERVER_READY
+                    "message": self.SERVER_READY,
+                    "backend": "faster_whisper"
                 }
             )
         )
@@ -729,6 +732,7 @@ class ServeClientFasterWhisper(ServeClientBase):
                         if time.time() - self.t_start > self.add_pause_thresh:
                             self.text.append('')
 
+                if not len(segments): continue
                 try:
                     self.websocket.send(
                         json.dumps({
@@ -781,6 +785,10 @@ class ServeClientFasterWhisper(ServeClientBase):
                 text_ = s.text
                 self.text.append(text_)
                 start, end = self.timestamp_offset + s.start, self.timestamp_offset + min(duration, s.end)
+
+                if start >= end: continue
+                if s.no_speech_prob > self.no_speech_thresh: continue
+
                 self.transcript.append(self.format_segment(start, end, text_))
                 
                 offset = min(duration, s.end)