tui-voice-mini-ml / api_server.py
Cyrax321's picture
Add OpenAI-compatible REST API server for robotics and home automation
769a950 verified
Raw History Blame Contribute Delete
1.42 kB
from fastapi import FastAPI, Response
from pydantic import BaseModel
import onnxruntime as ort
import numpy as np
import io, json
import soundfile as sf
import re
app = FastAPI(title="Tui Voice Mini Malayalam TTS API")
class SpeechRequest(BaseModel):
input: str
speed: float = 1.0
voice: int = 0
pitch_shift: float = 0.15
pitch_range: float = 0.85
# Load ONNX session and vocab
session = ort.InferenceSession("onnx/tui-voice-mini-ml.onnx")
cfg = json.load(open("config.json"))
vocab = cfg.get("vocab", {})
@app.post("/v1/audio/speech")
def generate_speech(req: SpeechRequest):
# Normalize & synthesize
from tui_voice import phonemes as P
tokens = P.tokenize_phonemes(P.text_to_phonemes(req.input))
token_ids = [vocab.get(t, 0) for t in tokens]
one = lambda v: np.array([v], dtype=np.float32)
audio, _ = session.run(None, {
"tokens": np.array([token_ids], dtype=np.int64),
"stress": np.zeros((1, len(token_ids)), dtype=np.int64),
"voice": np.array([req.voice], dtype=np.int64),
"speed": one(req.speed),
"pitch_shift": one(req.pitch_shift),
"pitch_range": one(req.pitch_range),
})
out = io.BytesIO()
sf.write(out, audio[0], 24000, format='WAV')
return Response(content=out.getvalue(), media_type="audio/wav")
if __name__ == "__main__":
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=8000)