no_speech_prob always returns 0.0.

[giaoyyds]

I'm using large-v3, and when I convert the audio to a numpy array and pass it to the model for transcription, the no_speech_prob returned is 0.0 every time, but with large-v2 there is a correct return.I can't fix this.Here's my sample code:

    def transcribe_audio(self, audio_numpy):
        try:
            model = WhisperModel("large-v3", device="cuda", compute_type="float16", local_files_only=False)

            result, info = model.transcribe(
                audio_numpy,
                initial_prompt="",
                language="en",
                task="transcribe",
                vad_filter=self.vad,
                vad_parameters={"threshold": 0.5}
            )

            all_segments = list(result)
            print(all_segments)

        except Exception as e:
            print(f"An error occurred during transcription: {e}")


    def send_audio_file(self, audio_file):
    
        print("do me.....")
        with open(audio_file, 'rb') as f:
            audio_data = f.read()
            audio_data = self. removewavhead(audio_data)

            for i in range(0, len(audio_data), 32000):
                chunk = audio_data[i:i + 32000]
                sf = soundfile.SoundFile(io.BytesIO(chunk), channels=2, endian="LITTLE", samplerate=8000, subtype="PCM_16", format="RAW")
                resampled_audio, _ = librosa.load(sf, sr=16000, dtype=np.float32)
                self.transcribe_audio(resampled_audio)
                time.sleep(0.1)

[MahmoudAshraf97]

I couldn't reproduce the issue on the sample audios included with this repo