add caching to sensevoice

app.py

@@ -240,13 +240,14 @@ def _transcribe_sense_cpu_stream(model_id: str, language: str, audio_path: str,
     with ProgressHook() as hook:
         diary = diarizer({"waveform": waveform, "sample_rate": sample_rate}, hook=hook)
     snippets = []
+    cache={}
     for turn, _, speaker in diary.itertracks(yield_label=True):
         start_ms, end_ms = int(turn.start*1000), int(turn.end*1000)
         segment = AudioSegment.from_file(audio_path)[start_ms:end_ms]
         with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
             segment.export(tmp.name, format="wav")
             try:
-                segs = model.generate(input=tmp.name, cache={}, language=language,
+                segs = model.generate(input=tmp.name, cache=cache, language=language,
                                         use_itn=enable_punct, batch_size_s=300)
             except Exception as e:
                 cprint(f'Error: {e}','red')
@@ -273,13 +274,14 @@ def _transcribe_sense_gpu_stream(model_id: str, language: str, audio_path: str,
     with ProgressHook() as hook:
         diary = diarizer({"waveform": waveform, "sample_rate": sample_rate}, hook=hook)
     snippets = []
+    cache = {}
     for turn, _, speaker in diary.itertracks(yield_label=True):
         start_ms, end_ms = int(turn.start*1000), int(turn.end*1000)
         segment = AudioSegment.from_file(audio_path)[start_ms:end_ms]
         with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
             segment.export(tmp.name, format="wav")
             try:
-                segs = model.generate(input=tmp.name, cache={}, language=language,
+                segs = model.generate(input=tmp.name, cache=cache, language=language,
                                         use_itn=enable_punct, batch_size_s=300)
             except Exception as e:
                 cprint(f'Error: {e}','red')