feat: Pass init_prompt to faster-whisper backend

whisperlivekit/core.py

@@ -115,7 +115,14 @@ def __init__(self, **kwargs):
                 )
 
             else:
-                self.asr, self.tokenizer = backend_factory(self.args)
+                self.asr, self.tokenizer = backend_factory(self.args.backend,
+                                                       lan=self.args.lan,
+                                                       modelsize=self.args.model,
+                                                       cache_dir=self.args.model_cache_dir,
+                                                       model_dir=self.args.model_dir,
+                                                       init_prompt=self.args.init_prompt)
+
+        if self.args.warmup_file:
             warmup_asr(self.asr, self.args.warmup_file) #for simulstreaming, warmup should be done in the online class not here
 
         if self.args.diarization:
@@ -165,4 +172,3 @@ def online_diarization_factory(args, diarization_backend):
         online = SortformerDiarizationOnline(shared_model=diarization_backend)
     return online
 
-

whisperlivekit/whisper_streaming_custom/backends.py

@@ -88,6 +88,10 @@ class FasterWhisperASR(ASRBase):
     """Uses faster-whisper as the backend."""
     sep = ""
 
+    def __init__(self, lan, modelsize=None, cache_dir=None, model_dir=None, logfile=sys.stderr, init_prompt=""):
+        super().__init__(lan, modelsize, cache_dir, model_dir, logfile)
+        self.init_prompt = init_prompt
+
     def load_model(self, modelsize=None, cache_dir=None, model_dir=None):
         from faster_whisper import WhisperModel
 
@@ -112,10 +116,12 @@ def load_model(self, modelsize=None, cache_dir=None, model_dir=None):
         return model
 
     def transcribe(self, audio: np.ndarray, init_prompt: str = "") -> list:
+        # faster-whisper's transcribe method has an `initial_prompt` argument
+        # which is perfect for this.
         segments, info = self.model.transcribe(
             audio,
             language=self.original_language,
-            initial_prompt=init_prompt,
+            initial_prompt=self.init_prompt or init_prompt,
             beam_size=5,
             word_timestamps=True,
             condition_on_previous_text=True,

whisperlivekit/whisper_streaming_custom/whisper_online.py

@@ -85,6 +85,7 @@ def backend_factory(args):
             lan=args.lan,
             cache_dir=getattr(args, 'model_cache_dir', None),
             model_dir=getattr(args, 'model_dir', None),
+            init_prompt=getattr(args, 'init_prompt', ""),  # <-- Here
         )
         e = time.time()
         logger.info(f"done. It took {round(e-t,2)} seconds.")