Realtime Diarization with microphone.

[u1846]

Can someone please help me setup this to feed in real live microphone audio to diarize at that time. I am seeing an function called mic_stream(). But nothing of what it is or how to setup .

Real-time streaming (e.g. microphone):

state = model.init_streaming_state()
for chunk in mic_stream(): # your audio source
for result in model.generate_stream(chunk, state=state, sample_rate=16000):
state = result.state
for seg in result.segments:
print(f"Speaker {seg.speaker}: {seg.start:.2f}s - {seg.end:.2f}s")

Somebody please help me setup this.

[pratikrath126]

Hey! The mlx-audio streaming API expects you to yield raw PCM chunks (16kHz, mono, 16-bit) from an iterator. You can set up mic_stream like this:

import sounddevice as sd
import numpy as np

def mic_stream(chunk_size=4096, sample_rate=16000):
    chunks = []
    def callback(indata, frames, time, status):
        pcm = (indata[:, 0] * 32767).astype(np.int16)
        chunks.append(pcm.tobytes())
    with sd.InputStream(samplerate=sample_rate, channels=1, dtype="float32", blocksize=chunk_size, callback=callback):
        while True:
            if chunks:
                yield chunks.pop(0)

Then use your loop:

state = model.init_streaming_state()
for chunk in mic_stream():
    for result in model.generate_stream(chunk, state=state, sample_rate=16000):
        state = result.state
        for seg in result.segments:
            print(f"Speaker {seg.speaker}: {seg.start:.2f}s - {seg.end:.2f}s")

If mic_stream() is already in the repo, check its signature – usually it yields bytes and you just iterate.

[u1846]

Thanks