app.py

import spaces
import gradio as gr
import torch
from TTS.api import TTS
import os
from pydub import AudioSegment
import re

os.makedirs("audio", exist_ok=True)

# Agree to Coqui TTS license
os.environ["COQUI_TOS_AGREED"] = "1"

# Auto-detect device
device = "cuda" if torch.cuda.is_available() else "cpu"
tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to(device)

# Function for long text voice cloning
@spaces.GPU(enable_queue=True)
def clone(text, audio):
    # Split input into sentences/phrases
    sentences = re.split(r'(?<=[.!?]) +', text)
    final_audio = AudioSegment.silent(duration=0)

    for i, chunk in enumerate(sentences):
        if not chunk.strip():
            continue
        temp_path = f"chunk_{i}.wav"
        tts.tts_to_file(
            text=chunk,
            speaker_wav=audio,
            language="en",
            file_path=temp_path
        )
        final_audio += AudioSegment.from_wav(temp_path)

    # Merge chunks into one file
    output_path = "./output.wav"
    final_audio.export(output_path, format="wav")
    return output_path

# UI
with gr.Blocks(theme=gr.themes.Soft(primary_hue="teal", secondary_hue="cyan", neutral_hue="slate")) as demo:
    
    # Custom CSS
    gr.HTML("""
    <style>
        body {
            background: linear-gradient(135deg, #0f172a, #1e293b);
            font-family: 'Inter', sans-serif;
            color: #f8fafc;
        }
        .gradio-container {
            max-width: 1200px;
            margin: auto;
        }
        .gr-block {
            background: #1e293b;
            border-radius: 16px;
            box-shadow: 0 8px 20px rgba(0,0,0,0.4);
            padding: 20px;
            transition: all 0.3s ease-in-out;
        }
        .gr-block:hover {
            box-shadow: 0 12px 28px rgba(0,0,0,0.6);
        }
        h1, h2, h3 {
            color: #06b6d4;
            font-weight: 700;
        }
        .gr-button.primary {
            background: linear-gradient(90deg, #06b6d4, #14b8a6);
            border: none;
            border-radius: 12px;
            font-weight: bold;
            transition: 0.3s;
        }
        .gr-button.primary:hover {
            background: linear-gradient(90deg, #14b8a6, #06b6d4);
            transform: scale(1.05);
        }
        .gr-textbox textarea {
            background: #0f172a !important;
            color: #f8fafc !important;
            border-radius: 12px;
            border: 1px solid #334155;
        }
        .gr-textbox textarea:focus {
            border-color: #06b6d4;
            outline: none !important;
            box-shadow: 0 0 10px rgba(6,182,212,0.5);
        }
        .gr-audio input, .gr-audio {
            border-radius: 12px !important;
            border: 1px solid #334155 !important;
            background: #0f172a !important;
        }
    </style>
    """)

    with gr.Row():
        with gr.Column(scale=1):
            gr.Markdown(
                """
                # 🎙️ Voice Clone Studio By Tahir Turk
                Clone any voice by uploading a short reference audio file  
                and typing what you want it to say.  
                **Powered by XTTS v2 — multilingual voice cloning.**
                """
            )

            text_input = gr.Textbox(
                label="Enter your text",
                placeholder="Type anything you'd like the cloned voice to say...",
                lines=6
            )
            audio_input = gr.Audio(
                type="filepath",
                label="Upload voice reference (WAV or MP3)"
            )
            submit_btn = gr.Button("✨ Generate Voice", variant="primary")

        with gr.Column(scale=1):
            output_audio = gr.Audio(type="filepath", label="🔊 Generated Voice Output")
            gr.Markdown(
                """
                ---
                ⚡ **Tips for Best Results**
                - Use a **clean, clear** reference audio (5–15 seconds works best).  
                - Long text will be split automatically for natural speech.  
                - You can generate **minutes of audio** now without cutoff.  
                ---
                """
            )

    with gr.Row():
        gr.Examples(
            examples=[
                ["Hey! It's me Dorthy, from the Wizard of Oz. Type in whatever you'd like me to say.", "./audio/Wizard-of-Oz-Dorthy.wav"],
                ["It's me Vito Corleone, from the Godfather. Type in whatever you'd like me to say.", "./audio/Godfather.wav"],
                ["Hey, it's me Paris Hilton. Type in whatever you'd like me to say.", "./audio/Paris-Hilton.mp3"],
                ["Hey, it's me Megan Fox from Transformers. Type in whatever you'd like me to say.", "./audio/Megan-Fox.mp3"],
                ["Hey there, it's me Jeff Goldblum. Type in whatever you'd like me to say.", "./audio/Jeff-Goldblum.mp3"],
                ["Hey there, it's me Heath Ledger as the Joker. Type in whatever you'd like me to say.", "./audio/Heath-Ledger.mp3"],
            ],
            inputs=[text_input, audio_input],
            outputs=[output_audio],
            label="🎭 Try with these sample voices"
        )

    submit_btn.click(fn=clone, inputs=[text_input, audio_input], outputs=output_audio)

demo.launch()