tipi/kyutai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Audio playback.

Browse Source
This commit is contained in:
Laurent 2025-07-02 15:49:31 +02:00
parent aa06a44fd4
commit 61206d78c8
1 changed files with 22 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
scripts/tts_mlx.py
View File

@ -19,6 +19,9 @@ import mlx.core as mx
import mlx.nn as nn import mlx.nn as nn
import sentencepiece import sentencepiece
import sphn import sphn
import time
import sounddevice as sd
from moshi_mlx.client_utils import make_log from moshi_mlx.client_utils import make_log
from moshi_mlx import models from moshi_mlx import models
@ -126,11 +129,25 @@ def main():
# We are processing frames one by one, although we could group them to improve speed. # We are processing frames one by one, although we could group them to improve speed.
_pcm = tts_model.mimi.decode_step(frame) _pcm = tts_model.mimi.decode_step(frame)
wav_frames.append(_pcm) wav_frames.append(_pcm)
wavs = mx.concat(wav_frames, axis=-1) if args.out == "-":
end_step = result.end_steps[0] cnt = [0]
wav_length = int((mimi.sample_rate * (end_step + tts_model.final_padding) / mimi.frame_rate)) def audio_callback(outdata, _a, _b, _c):
wav = wavs[0, :, :wav_length] if cnt[0] < len(wav_frames):
sphn.write_wav(args.out, np.array(mx.clip(wav, -1, 1)), mimi.sample_rate) outdata[:, 0] = wav_frames[cnt[0]][0, 0]
cnt[0] += 1
else:
outdata[:] = 0
with sd.OutputStream(samplerate=mimi.sample_rate,
blocksize=1920,
channels=1,
callback=audio_callback):
time.sleep(10)
else:
wavs = mx.concat(wav_frames, axis=-1)
end_step = result.end_steps[0]
wav_length = int((mimi.sample_rate * (end_step + tts_model.final_padding) / mimi.frame_rate))
wav = np.array(mx.clip(wavs[0, :, :wav_length], -1, 1))
sphn.write_wav(args.out, wav, mimi.sample_rate)
if __name__ == "__main__": if __name__ == "__main__":