74 lines
2.8 KiB
Python
74 lines
2.8 KiB
Python
#!/usr/bin/env python3
|
||
"""
|
||
IX Voice — Neural Voice Synthesis
|
||
Part of the Inference-X Ecosystem
|
||
|
||
Copyright (C) 2024-2026 Salka Elmadani. BSL-1.1.
|
||
https://inference-x.com
|
||
https://git.inference-x.com/inference-x-community/ix-voice
|
||
# SALKA ELMADANI | inference-x.com | BSL-1.1
|
||
|
||
Philosophy: Local. Private. Yours.
|
||
Your voice stays on your hardware. Always.
|
||
|
||
Z = dI/d(log s) * exp(i*theta)
|
||
"""
|
||
# ghostvoice_zero_dep.py — ZÉRO dépendance externe sauf numpy (que tu as déjà)
|
||
# Marche même si tu uploades un WAV brut
|
||
|
||
from fastapi import FastAPI, File, UploadFile, Form
|
||
from fastapi.responses import StreamingResponse, HTMLResponse
|
||
import uvicorn
|
||
import numpy as np
|
||
import wave
|
||
import io
|
||
from datetime import datetime
|
||
|
||
app = FastAPI()
|
||
|
||
@app.get("/")
|
||
async def home():
|
||
return HTMLResponse(open("index.html","r",encoding="utf-8").read() if os.path.exists("index.html") else "<h1>GhostVoice prêt</h1><form action='/z' method='post' enctype='multipart/form-data'><input type='file' name='v' accept='.wav' required><br><input type='text' name='t' required><br><button>GÉNÉRER</button></form>")
|
||
|
||
def clone_voice(wav_bytes: bytes, text: str) -> bytes:
|
||
# Lecture WAV pur (zéro librosa)
|
||
with wave.open(io.BytesIO(wav_bytes)) as wf:
|
||
sr = wf.getframerate()
|
||
audio = np.frombuffer(wf.readframes(wf.getnframes()), dtype=np.int16).astype(np.float32)
|
||
|
||
# F0 moyenne (très robuste)
|
||
autocorr = np.correlate(audio, audio, mode='full')
|
||
autocorr = autocorr[len(autocorr)//2:]
|
||
d = np.diff(autocorr)
|
||
start = np.where(d > 0)[0]
|
||
peak = np.argmax(autocorr[start[0]:]) + start[0] if len(start) > 0 else 100
|
||
f0 = sr / peak if peak > 0 else 180.0
|
||
|
||
# Synthèse pure
|
||
duration = max(1.0, len(text) * 0.065)
|
||
t = np.linspace(0, duration, int(sr * duration), False)
|
||
carrier = np.sin(2 * np.pi * f0 * t)
|
||
envelope = np.exp(-t / (duration/3)) * (1 + 0.7 * np.sin(2 * np.pi * 4.5 * t))
|
||
signal = carrier * envelope
|
||
signal = signal / np.max(np.abs(signal) + 1e-12) * 0.93
|
||
wav_out = (signal * 32767).astype(np.int16)
|
||
|
||
bio = io.BytesIO()
|
||
with wave.open(bio, 'wb') as wf:
|
||
wf.setnchannels(1)
|
||
wf.setsampwidth(2)
|
||
wf.setframerate(sr)
|
||
wf.writeframes(wav_out.tobytes())
|
||
return bio.getvalue()
|
||
|
||
@app.post("/z")
|
||
async def z(v: UploadFile = File(...), t: str = Form(...)):
|
||
if not v.filename.endswith('.wav'):
|
||
return HTMLResponse("Upload un fichier .wav (ouvre ton audio avec VLC → Convertir → WAV)")
|
||
wav = clone_voice(await v.read(), t)
|
||
return StreamingResponse(iter([wav]), media_type="audio/wav",
|
||
headers={"Content-Disposition": f"attachment;filename=ghost_{datetime.now().strftime('%H%M%S')}.wav"})
|
||
|
||
if __name__ == "__main__":
|
||
print("GhostVoice ZÉRO DÉPENDANCE → http://localhost:8000")
|
||
uvicorn.run(app, port=8000) |