app.py

import spaces
import os
import torch
import soundfile as sf
import logging
import gradio as gr
import librosa
import numpy as np
from datetime import datetime
import spaces
from transformers import AutoTokenizer, AutoModelForCausalLM

# ------------------------------
# Logging
# ------------------------------
logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s - %(levelname)s - %(message)s"
)

# ------------------------------
# Global Model
# ------------------------------
MODEL = None
TOKENIZER = None
MODEL_ID = "rahul7star/mir-TTS"

# ------------------------------
# Helper Functions
# ------------------------------
def load_model():
    """Lazy load model and tokenizer."""
    global MODEL, TOKENIZER
    if MODEL is None or TOKENIZER is None:
        logging.info(f"Loading model: {MODEL_ID}")
        MODEL = AutoModelForCausalLM.from_pretrained(MODEL_ID).cuda()
        TOKENIZER = AutoTokenizer.from_pretrained(MODEL_ID)
        logging.info("Model loaded on GPU")
    return MODEL, TOKENIZER

def validate_audio_input(audio_path):
    """Validate and preprocess audio input."""
    if not audio_path or not os.path.exists(audio_path):
        raise ValueError("Audio file not found")
    audio, sr = librosa.load(audio_path, sr=None, duration=30)
    if len(audio) == 0:
        raise ValueError("Audio is empty")
    # Minimum 0.5 seconds
    if len(audio) < int(0.5 * sr):
        raise ValueError("Audio too short, must be >=0.5s")
    # Resample to 16kHz
    if sr != 16000:
        audio = librosa.resample(audio, orig_sr=sr, target_sr=16000)
        sr = 16000
    # Normalize
    audio = audio / np.max(np.abs(audio))
    # Save temp file
    temp_dir = "/tmp" if os.path.exists("/tmp") else "."
    temp_path = os.path.join(temp_dir, f"processed_{os.path.basename(audio_path)}")
    sf.write(temp_path, audio, samplerate=sr)
    return temp_path, sr

# ------------------------------
# Core Generation Function
# ------------------------------
@spaces.GPU()
def generate_speech(text, prompt_audio_path):
    """Generate speech from text with reference audio."""
    try:
        model, tokenizer = load_model()

        if not text or not text.strip():
            raise ValueError("Text is empty")

        # Preprocess audio
        processed_audio, sr = validate_audio_input(prompt_audio_path)

        # Encode audio as context tokens
        audio_input_ids = tokenizer.apply_chat_template(
            [{"role": "user", "content": "Encode audio context"}],
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            return_tensors="pt"
        ).to(model.device)

        # Simple text generation using tokens
        text_input_ids = tokenizer.apply_chat_template(
            [{"role": "user", "content": text}],
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            return_tensors="pt"
        ).to(model.device)

        outputs = model.generate(
            **text_input_ids,
            max_new_tokens=512
        )
        generated_text = tokenizer.decode(outputs[0][text_input_ids["input_ids"].shape[-1]:])

        # For demo, return generated text as placeholder audio
        # You can integrate your TTS codec here
        dummy_audio = np.random.rand(sr * 2).astype("float32") * 0.01

        # Cleanup
        if os.path.exists(processed_audio):
            os.remove(processed_audio)

        return dummy_audio, 48000

    except Exception as e:
        logging.error(f"Generation error: {e}")
        raise e

# ------------------------------
# Gradio Interface
# ------------------------------
def voice_clone_interface(text, prompt_audio_upload, prompt_audio_record):
    """Interface callback for voice cloning."""
    try:
        prompt_audio = prompt_audio_upload or prompt_audio_record
        if not prompt_audio:
            return None, "Upload or record reference audio first"
        if not text.strip():
            return None, "Enter text to synthesize"

        audio, sr = generate_speech(text, prompt_audio)

        # Save output
        os.makedirs("outputs", exist_ok=True)
        timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
        output_path = f"outputs/mir_tts_{timestamp}.wav"
        sf.write(output_path, audio, samplerate=sr)

        return output_path, "Generation successful!"

    except Exception as e:
        logging.error(f"Voice clone error: {e}")
        return None, f"Error: {e}"

def build_interface():
    """Build Gradio interface."""
    with gr.Blocks(title="MiraTTS Voice Cloning") as demo:
        gr.HTML("<h1 style='text-align:center;color:#2563eb;'>MiraTTS Voice Cloning</h1>")
        with gr.Row():
            with gr.Column():
                gr.Markdown("### Reference Audio")
                prompt_upload = gr.Audio(sources="upload", type="filepath")
                prompt_record = gr.Audio(sources="microphone", type="filepath")
            with gr.Column():
                gr.Markdown("### Text Input")
                text_input = gr.Textbox(
                    placeholder="Enter text...",
                    lines=4,
                    value="Hello! This is a demonstration of MiraTTS"
                )
                generate_btn = gr.Button("Generate Speech", variant="primary")

        with gr.Row():
            output_audio = gr.Audio(label="Generated Speech", type="filepath", autoplay=True)
            status_text = gr.Textbox(label="Status", interactive=False)

        generate_btn.click(
            voice_clone_interface,
            inputs=[text_input, prompt_upload, prompt_record],
            outputs=[output_audio, status_text]
        )

        def clear_all():
            return None, None, "", None, "Ready for new generation"
        clear_btn = gr.Button("Clear All", variant="secondary")
        clear_btn.click(
            clear_all,
            outputs=[prompt_upload, prompt_record, text_input, output_audio, status_text]
        )

    return demo

if __name__ == "__main__":
    demo = build_interface()
    demo.launch(server_name="0.0.0.0", server_port=7860, share=False)