app.py

import gradio as gr
import torch
from transformers import pipeline
from langdetect import detect

# Load the Coqui XTTS model
tts = pipeline("text-to-speech", model="coqui/XTTS-v2", device=0 if torch.cuda.is_available() else -1)

# Helper function to clone voice and generate speech
def clone_and_generate(audio, text_prompt, language):
    if audio is None or text_prompt.strip() == "":
        return "Please provide both audio input and text prompt.", None
    
    # Check if language is supported
    supported_languages = {"english": "en", "hindi": "hi"}
    if language not in supported_languages:
        return f"Language {language} not supported yet.", None

    # Convert text to the target language (if needed)
    if detect(text_prompt) != supported_languages[language]:
        # For now, we assume text is already in the desired language
        pass

    # Generate speech
    try:
        result = tts(text=text_prompt, speaker=audio)
        return "Speech generated successfully!", result["audio"]
    except Exception as e:
        return f"Error: {str(e)}", None

# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("## 🎤 Voice Cloning & Text-to-Speech with Language Translation")
    
    with gr.Row():
        with gr.Column():
            audio_input = gr.Audio(source="microphone", type="filepath", label="🎙️ Record or Upload Voice")
            text_input = gr.Textbox(label="📝 Enter Text to Generate Speech")
            language_input = gr.Dropdown(choices=["english", "hindi"], value="english", label="🌐 Select Language")

        with gr.Column():
            output_message = gr.Textbox(label="📢 Status")
            output_audio = gr.Audio(label="🔊 Generated Speech")

    generate_button = gr.Button("🚀 Generate Speech")
    generate_button.click(clone_and_generate, inputs=[audio_input, text_input, language_input], outputs=[output_message, output_audio])

# Launch the app
demo.launch()