app.py

import gradio as gr
import torch
import torchaudio
import tempfile
import os
import warnings
from contextlib import contextmanager

warnings.filterwarnings("ignore")
os.environ["COQUI_TOS_AGREED"] = "1"
print("🚀 Starting Voice Cloning Studio...")

@contextmanager
def patch_torch_load():
    original_load = torch.load
    def patched_load(f, *args, **kwargs):
        kwargs['weights_only'] = False
        return original_load(f, *args, **kwargs)
    torch.load = patched_load
    try:
        yield
    finally:
        torch.load = original_load

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
TTS_MODEL = None
WHISPER_MODEL = None
MODEL_STATUS = "Not Loaded"

def load_xtts_manual():
    global TTS_MODEL, MODEL_STATUS
    if TTS_MODEL is not None:
        return True
    try:
        with patch_torch_load():
            from TTS.api import TTS
            print("📦 Loading XTTS...")
            TTS_MODEL = TTS(
                model_name="tts_models/multilingual/multi-dataset/xtts_v2",
                progress_bar=True,
                gpu=(DEVICE == "cuda")
            )
            MODEL_STATUS = "XTTS-v2 Ready"
            print("✅ XTTS loaded!")
            return True
    except Exception as e:
        print(f"❌ XTTS loading failed: {e}")
        MODEL_STATUS = f"Manual Failed: {str(e)}"
        return False

def load_whisper():
    global WHISPER_MODEL
    if WHISPER_MODEL is not None:
        return True
    try:
        import whisper
        WHISPER_MODEL = whisper.load_model("base")
        print("✅ Whisper loaded!")
        return True
    except Exception as e:
        print(f"❌ Whisper failed: {e}")
        return False

def voice_to_voice_clone(reference_audio, input_audio, language="en"):
    """
    Main voice cloning function - this will be called by both UI and API
    """
    try:
        print(f"🎭 Voice cloning request: {language}")
        print(f"📁 Reference: {reference_audio}")
        print(f"📁 Input: {input_audio}")
        
        if not reference_audio or not input_audio:
            return None, "❌ Please upload both reference and input audio files!"
        
        # Load XTTS model
        if not load_xtts_manual():
            return None, f"❌ XTTS loading failed!\nStatus: {MODEL_STATUS}"
        
        # Load Whisper for transcription
        load_whisper()
        
        # Extract text from input audio
        extracted_text = "Voice cloning demonstration."
        if WHISPER_MODEL:
            try:
                result = WHISPER_MODEL.transcribe(input_audio)
                text = result.get("text", "").strip()
                if text and len(text) > 3:
                    extracted_text = text
                print(f"✅ Extracted: '{extracted_text[:100]}...'")
            except Exception as e:
                print(f"⚠️ Whisper error: {e}")
        
        # Generate cloned voice
        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_file:
            output_path = tmp_file.name
        
        print(f"🔄 Generating voice clone...")
        with patch_torch_load():
            TTS_MODEL.tts_to_file(
                text=extracted_text,
                speaker_wav=reference_audio,
                language=language,
                file_path=output_path
            )
        
        # Verify output
        if os.path.exists(output_path) and os.path.getsize(output_path) > 0:
            success_message = f"""✅ VOICE-TO-VOICE CLONING SUCCESS!
📝 Content: '{extracted_text[:150]}...'
🎭 Device: {DEVICE}
🔧 Status: {MODEL_STATUS}
📊 Output size: {os.path.getsize(output_path)} bytes
"""
            print("✅ Voice cloning completed successfully!")
            return output_path, success_message
        else:
            return None, "❌ Generated audio file is empty!"
            
    except Exception as e:
        error_msg = f"❌ Voice cloning error: {str(e)}\nModel: {MODEL_STATUS}"
        print(error_msg)
        return None, error_msg

# FIXED: Use gr.Interface instead of gr.Blocks for proper API exposure
interface = gr.Interface(
    fn=voice_to_voice_clone,
    inputs=[
        gr.Audio(
            label="🎤 Reference Audio (Voice to Clone)",
            type="filepath",
            sources=["upload"]
        ),
        gr.Audio(
            label="🎵 Input Audio (Content to Transform)", 
            type="filepath",
            sources=["upload"]
        ),
        gr.Dropdown(
            choices=[
                "en", "es", "fr", "de", "it", "pt", "pl", "tr", "ru", "nl", 
                "cs", "ar", "zh", "ja", "ko", "hi", "uk", "vi", "ro", "el",
                "he", "fi", "hu", "sv", "ca", "id", "ms", "bg", "sk", "da",
                "no", "lt", "hr", "sr", "sl", "et", "lv", "fil", "bn", "ta",
                "te", "ur", "fa", "th"
            ],
            value="en",
            label="🌍 Language"
        )
    ],
    outputs=[
        gr.Audio(label="🎉 Cloned Voice Result"),
        gr.Textbox(label="📋 Status", lines=8)
    ],
    title="🎭 REAL Voice Cloning Studio",
    description="Transform any voice into any other voice using XTTS-v2 and Whisper AI models. Upload reference audio and input audio to get started.",
    theme=gr.themes.Soft(),
    allow_flagging="never",
    api_name="voice_to_voice_clone"  # CRITICAL: This creates the API endpoint
)

if __name__ == "__main__":
    print("🌐 Launching Voice Cloning Studio...")
    interface.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,
        show_api=True,  # Shows API documentation
        debug=True
    )