mlx-whisper real time audio

mlx_whisper_realtime.py

# /// script
# dependencies = [
#   "SpeechRecognition",
#   "mlx-whisper",
#   "pyaudio",
# ]
# ///

import speech_recognition as sr
import numpy as np
import mlx_whisper

r = sr.Recognizer()
mic = sr.Microphone(sample_rate=16000)

print("Listening...")

try:
    with mic as source:
        r.adjust_for_ambient_noise(source)
        while True:
            audio = r.listen(source)
            # Convert audio to numpy array
            audio_data = np.frombuffer(audio.get_raw_data(), dtype=np.int16).astype(np.float32) / 32768.0
            
            # Process audio with Apple MLXWhisper model
            result = mlx_whisper.transcribe(audio_data, path_or_hf_repo="mlx-community/whisper-large-v3-turbo")["text"]
            
            # Print the transcribed text
            print(result)

except KeyboardInterrupt:
    print("Stopped listening.")

With that comment you can run this using:

uv run mlx_whisper_realtime.py

🔝

on a clean install I needed to first install Port Audio with brew install portaudio