app.py

import os
import torch
import torchaudio
import threading
import gradio as gr
import requests
from fastapi import FastAPI
from fastapi.responses import FileResponse
from huggingface_hub import snapshot_download, hf_hub_download
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts
import uvicorn

# ===== MODEL SETUP =====
checkpoint_dir = "model/"
repo_id = "capleaf/viXTTS"
os.makedirs(checkpoint_dir, exist_ok=True)
api_app = FastAPI()

required_files = ["model.pth", "config.json", "vocab.json", "speakers_xtts.pth"]
if not all(f in os.listdir(checkpoint_dir) for f in required_files):
    snapshot_download(repo_id=repo_id, local_dir=checkpoint_dir)
    hf_hub_download("coqui/XTTS-v2", "speakers_xtts.pth", local_dir=checkpoint_dir)

config = XttsConfig()
config.load_json(os.path.join(checkpoint_dir, "config.json"))
MODEL = Xtts.init_from_config(config)
MODEL.load_checkpoint(config, checkpoint_dir=checkpoint_dir, use_deepspeed=False)

# CPU-only
MODEL.cpu()
MODEL.gpt.float()
torch.set_num_threads(4)
torch.backends.mkldnn.enabled = True

LANGS = ["vi", "en", "zh-cn", "ja", "ko"]

# ===== TTS FUNCTION =====
DEFAULT_REF = "model/samples/nu-luu-loat.wav"  # file mẫu mặc định

def tts_fn(text, language):
    ref_audio = DEFAULT_REF
    import os
    print(">>> Server-side ref_audio path:", os.path.abspath(ref_audio))
    print(">>> Exists:", os.path.exists(ref_audio))

    gpt_latent, spk_embed = MODEL.get_conditioning_latents(
        audio_path=ref_audio,
        gpt_cond_len=18,
        gpt_cond_chunk_len=4,
        max_ref_length=50
    )
    out = MODEL.inference(
        text=text,
        language=language,
        gpt_cond_latent=gpt_latent,  
        speaker_embedding=spk_embed,
        temperature=0.65,
        repetition_penalty=2.5,
        enable_text_splitting=True
    )
    wav = torch.tensor(out["wav"]).unsqueeze(0)
    out_path = "output.wav"
    torchaudio.save(out_path, wav, 24000)
    print(">>> Generated wav path:", os.path.abspath(out_path))
    return out_path

# ===== FASTAPI SERVER =====
@api_app.post("/api/speak")
def speak_api(text: str, language: str = "vi"):
    try:
        path = tts_fn(text, language)
        return FileResponse(path, media_type="audio/wav")
    except Exception as e:
        return {"error": str(e)}

# ===== GRADIO CLIENT =====
def gradio_client(text, language):
    try:
        r = requests.post(
            "http://127.0.0.1:8000/api/speak",
            params={"text": text, "language": language}
        )
        if r.status_code == 200:
            with open("voice.wav", "wb") as f:
                f.write(r.content)
            return "voice.wav", "✅ Hoàn tất!"
        else:
            return None, f"❌ Lỗi API: {r.status_code}"
    except Exception as e:
        return None, f"❌ Lỗi: {str(e)}"

# ===== GRADIO UI =====
with gr.Blocks(title="ViXTTS - Gradio + API") as demo:
    gr.Markdown("## 🎙️ Vietnamese TTS - CPU (Spaces HuggingFace)")

    with gr.Row():
        with gr.Column(scale=1):
            text_in = gr.Textbox(label="Văn bản", value="Xin chào!", lines=4)
            lang_dd = gr.Dropdown(label="Ngôn ngữ", choices=LANGS, value="vi")
            btn = gr.Button("🎧 Tạo giọng")

        with gr.Column(scale=1):
            audio_out = gr.Audio(label="Kết quả", autoplay=True)
            info_out = gr.Textbox(label="Trạng thái", interactive=False)

    btn.click(gradio_client, inputs=[text_in, lang_dd], outputs=[audio_out, info_out])

# ===== CHẠY SONG SONG API + GRADIO =====
if __name__ == "__main__":
    threading.Thread(target=lambda: uvicorn.run(api_app, host="0.0.0.0", port=8000), daemon=True).start()
    demo.launch(server_name="0.0.0.0", server_port=7860)