app.py

import io

from fastapi import FastAPI, HTTPException, Query
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import StreamingResponse
from kokoro_onnx import Kokoro
import soundfile as sf

MODEL_PATH = "kokoro-v1.0.onnx"
VOICES_PATH = "voices-v1.0.bin"

VOICES = {
    "en": ["af_heart", "am_puck"],
    "fr": ["ff_siwis"],
}
LANG_CODES = {"en": "en-us", "fr": "fr-fr"}

app = FastAPI(title="Speakeo TTS", description="API Kokoro TTS (EN/FR)")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # à restreindre au domaine de Speakeo plus tard
    allow_methods=["*"],
    allow_headers=["*"],
)

_kokoro: Kokoro | None = None


def get_kokoro() -> Kokoro:
    global _kokoro
    if _kokoro is None:
        _kokoro = Kokoro(MODEL_PATH, VOICES_PATH)
    return _kokoro


@app.get("/")
def root():
    return {"status": "ok", "service": "speakeo-tts", "languages": list(VOICES)}


@app.get("/voices")
def voices():
    return VOICES


@app.get("/tts")
def tts(
    text: str = Query(..., min_length=1, max_length=2000),
    lang: str = Query("en", pattern="^(en|fr)$"),
    voice: str | None = None,
    speed: float = Query(1.0, ge=0.5, le=2.0),
):
    if voice is None:
        voice = VOICES[lang][0]
    elif voice not in VOICES[lang]:
        raise HTTPException(400, f"Voix invalide pour '{lang}'. Options : {VOICES[lang]}")

    samples, sample_rate = get_kokoro().create(
        text, voice=voice, speed=speed, lang=LANG_CODES[lang]
    )

    buf = io.BytesIO()
    sf.write(buf, samples, sample_rate, format="WAV")
    buf.seek(0)
    return StreamingResponse(buf, media_type="audio/wav")