app.py

import gradio as gr
import torch
from transformers import AutoModelForTextToWaveform, AutoProcessor

# Load model and processor
model_name = "hexgrad/Kokoro-82M"
processor = AutoProcessor.from_pretrained(model_name)
model = AutoModelForTextToWaveform.from_pretrained(model_name, torch_dtype=torch.float16)

# Move to GPU if available
device = "cuda" if torch.cuda.is_available() else "cpu"
model = model.to(device)

def text_to_audio(text, speed=1.0):
    """Convert text to audio using Kokoro model"""
    # Process the input text
    inputs = processor(text=text, return_tensors="pt")
    inputs = {k: v.to(device) for k, v in inputs.items()}
    
    # Set generation parameters
    gen_kwargs = {
        "do_sample": True,
        "temperature": 0.7,
        "length_penalty": 1.0,
        "repetition_penalty": 2.0,
        "top_p": 0.9,
    }
    
    # Generate waveform
    with torch.no_grad():
        waveform = model.generate(**inputs, **gen_kwargs).cpu().numpy()[0]
    
    # Create a sample rate (typical for audio is 24000)
    sample_rate = 24000
    
    # Apply speed factor if needed
    if speed != 1.0:
        import numpy as np
        import librosa
        waveform = librosa.effects.time_stretch(waveform.astype(np.float32), rate=speed)
    
    return sample_rate, waveform

# Create Gradio interface
with gr.Blocks(title="Kokoro Text-to-Audio") as app:
    gr.Markdown("# 🎵 Kokoro Text-to-Audio Converter")
    gr.Markdown("Convert text to speech using hexgrad/Kokoro-82M model")
    
    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(
                label="Enter your text",
                placeholder="Type something to convert to audio...",
                lines=5
            )
            speed_slider = gr.Slider(
                minimum=0.5,
                maximum=1.5,
                value=1.0,
                step=0.1,
                label="Speech Speed"
            )
            submit_btn = gr.Button("Generate Audio")
        
        with gr.Column():
            audio_output = gr.Audio(label="Generated Audio", type="numpy")
    
    submit_btn.click(
        fn=text_to_audio,
        inputs=[text_input, speed_slider],
        outputs=[audio_output]
    )
    
    gr.Markdown("### Usage Tips")
    gr.Markdown("- For best results, keep your text reasonably short")
    gr.Markdown("- Adjust the speed slider to modify the pace of speech")
    gr.Markdown("- The model may take a moment to load on first use")

# Launch the app
if __name__ == "__main__":
    app.launch()