app.py

from fastapi import FastAPI, UploadFile, File
from fastapi.responses import FileResponse
from huggingface_hub import InferenceClient
import os

# Khởi tạo FastAPI
app = FastAPI()

# Lấy API key từ biến môi trường (Không lộ API key)
HF_API_KEY = os.getenv("HF_API_KEY")
if not HF_API_KEY:
    raise ValueError("Missing HF_API_KEY. Add it to Hugging Face Spaces Secrets.")

client = InferenceClient(token=HF_API_KEY)

# Thư mục lưu file tạm thời
TEMP_DIR = "temp"
os.makedirs(TEMP_DIR, exist_ok=True)

# Mô hình TTS & STT tối ưu CPU
STT_MODEL = "openai/whisper-tiny.en"
TTS_MODEL = "facebook/mms-tts-eng"

# API TTS: Chuyển văn bản thành giọng nói
@app.post("/tts")
async def text_to_speech(text: str):
    output_path = os.path.join(TEMP_DIR, "output.wav")

    try:
        audio = client.text_to_speech(model=TTS_MODEL, inputs=text)
        with open(output_path, "wb") as f:
            f.write(audio)
        return FileResponse(output_path, media_type="audio/wav", filename="output.wav")
    except Exception as e:
        return {"error": str(e)}

# API STT: Chuyển giọng nói thành văn bản
@app.post("/stt")
async def speech_to_text(file: UploadFile = File(...)):
    try:
        audio_data = await file.read()
        text = client.automatic_speech_recognition(model=STT_MODEL, data=audio_data)
        return {"text": text}
    except Exception as e:
        return {"error": str(e)}

# Kiểm tra API đang chạy
@app.get("/")
async def root():
    return {"message": "TTS & STT API is running!"}