Hacker-Software/ARIA-AGENT
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases 236 Wiki Activity

fix(voxtral): Audio als temp-WAV-Pfad an Processor (statt rohem Array)

Browse Source 
VoxtralProcessor.apply_transcription_request verlangt bei rohen Arrays ein 'format'. Fix: Buffer in ein temp-WAV (PCM_16, 16kHz) schreiben und den Pfad uebergeben — Processor liest Format+Samplerate selbst. Temp-Datei wird nach dem Transkribieren geloescht.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
duffyduck
2026-08-15 11:29:51 +02:00
co-authored by Claude Opus 4.8
parent fff2e7df34
commit 3324d39d50
1 changed files with 25 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
xtts/voxtral/bridge.py
+25 -10
View File
@@ -29,11 +29,13 @@ import base64
import json import json
import logging import logging
import os import os
import tempfile
import time import time
from dataclasses import dataclass, field from dataclasses import dataclass, field
from typing import Optional from typing import Optional
import numpy as np import numpy as np
import soundfile as sf
import websockets import websockets
logging.basicConfig( logging.basicConfig(
@@ -102,20 +104,33 @@ class VoxtralRunner:
logger.info("Voxtral geladen in %.1fs", time.time() - t0) logger.info("Voxtral geladen in %.1fs", time.time() - t0)
def _transcribe_blocking(self, audio_f32: np.ndarray, language: str) -> str: def _transcribe_blocking(self, audio_f32: np.ndarray, language: str) -> str:
# ⚠️ VERIFY: exakte Voxtral-Transformers-API gegen die HF-Modelcard.
import torch import torch
proc, model = self.processor, self.model proc, model = self.processor, self.model
if proc is None or model is None or audio_f32.size == 0: if proc is None or model is None or audio_f32.size == 0:
return "" return ""
inputs = proc.apply_transcription_request( # VoxtralProcessor verlangt bei rohen Arrays ein 'format'. Robuster:
language=language, audio=audio_f32, model_id=VOXTRAL_MODEL, sampling_rate=16000, # in ein temp-WAV schreiben und den PFAD uebergeben — der Processor liest
) # Format + Samplerate selbst, kein 'format'-Argument noetig.
inputs = inputs.to(VOXTRAL_DEVICE, dtype=torch.bfloat16) wav_path = None
with torch.no_grad(): try:
outputs = model.generate(**inputs, max_new_tokens=512) with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tf:
trimmed = outputs[:, inputs.input_ids.shape[1]:] wav_path = tf.name
text = proc.batch_decode(trimmed, skip_special_tokens=True) sf.write(wav_path, audio_f32, 16000, subtype="PCM_16")
return (text[0] if text else "").strip() inputs = proc.apply_transcription_request(
language=language, audio=wav_path, model_id=VOXTRAL_MODEL,
)
inputs = inputs.to(VOXTRAL_DEVICE, dtype=torch.bfloat16)
with torch.no_grad():
outputs = model.generate(**inputs, max_new_tokens=512)
trimmed = outputs[:, inputs.input_ids.shape[1]:]
text = proc.batch_decode(trimmed, skip_special_tokens=True)
return (text[0] if text else "").strip()
finally:
if wav_path:
try:
os.unlink(wav_path)
except Exception:
pass
async def transcribe(self, audio_f32: np.ndarray, language: str) -> str: async def transcribe(self, audio_f32: np.ndarray, language: str) -> str:
loop = asyncio.get_running_loop() loop = asyncio.get_running_loop()