app.py

# This is a Gradio app that generates audio from text using the medium-sized AudioGen model.
import gradio as gr
import numpy as np
from transformers_js_py import AutoModelForAudioGeneration, AutoTokenizer

# Load the medium-sized AudioGen model and tokenizer
model_name = "facebook/audiogen-medium"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForAudioGeneration.from_pretrained(model_name)

# Define a function that takes text and generation parameters, and generates audio.
def text_to_audio(text, duration, use_sampling=True, top_k=250, top_p=0.0, temperature=1.0, cfg_coef=3.0):
    # Tokenize the input text
    inputs = tokenizer(text, return_tensors="pt")
    
    # Generate audio from the tokenized input with the specified parameters
    audio = model.generate(
        **inputs,
        max_length=int(duration * 44100),
        use_sampling=use_sampling,
        top_k=top_k,
        top_p=top_p,
        temperature=temperature,
        cfg_coef=cfg_coef
    )
    
    # Convert the generated audio to a format suitable for Gradio
    sr = 44100  # Sample rate
    audio = audio.squeeze().numpy().astype(np.float32)
    
    return (sr, audio)

# Create a Gradio interface that takes a textbox input, a slider for duration, and sliders for generation parameters.
with gr.Blocks() as demo:
    text_input = gr.Textbox(label="Text Input")
    duration_slider = gr.Slider(minimum=3, maximum=15, step=1, label="Duration (seconds)")
    use_sampling_checkbox = gr.Checkbox(label="Use Sampling", value=True)
    top_k_slider = gr.Slider(minimum=1, maximum=1000, step=1, label="Top K", value=250)
    top_p_slider = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label="Top P", value=0.0)
    temperature_slider = gr.Slider(minimum=0.1, maximum=5.0, step=0.1, label="Temperature", value=1.0)
    cfg_coef_slider = gr.Slider(minimum=0.1, maximum=10.0, step=0.1, label="CFG Coefficient", value=3.0)
    audio_output = gr.Audio(label="Generated Audio")
    
    # Define the function to update the audio output
    def update_audio(text, duration, use_sampling, top_k, top_p, temperature, cfg_coef):
        return text_to_audio(text, duration, use_sampling, top_k, top_p, temperature, cfg_coef)
    
    # Set up the change events for the input components
    text_input.change(fn=update_audio, inputs=[text_input, duration_slider, use_sampling_checkbox, top_k_slider, top_p_slider, temperature_slider, cfg_coef_slider], outputs=audio_output)
    duration_slider.change(fn=update_audio, inputs=[text_input, duration_slider, use_sampling_checkbox, top_k_slider, top_p_slider, temperature_slider, cfg_coef_slider], outputs=audio_output)
    use_sampling_checkbox.change(fn=update_audio, inputs=[text_input, duration_slider, use_sampling_checkbox, top_k_slider, top_p_slider, temperature_slider, cfg_coef_slider], outputs=audio_output)
    top_k_slider.change(fn=update_audio, inputs=[text_input, duration_slider, use_sampling_checkbox, top_k_slider, top_p_slider, temperature_slider, cfg_coef_slider], outputs=audio_output)
    top_p_slider.change(fn=update_audio, inputs=[text_input, duration_slider, use_sampling_checkbox, top_k_slider, top_p_slider, temperature_slider, cfg_coef_slider], outputs=audio_output)
    temperature_slider.change(fn=update_audio, inputs=[text_input, duration_slider, use_sampling_checkbox, top_k_slider, top_p_slider, temperature_slider, cfg_coef_slider], outputs=audio_output)
    cfg_coef_slider.change(fn=update_audio, inputs=[text_input, duration_slider, use_sampling_checkbox, top_k_slider, top_p_slider, temperature_slider, cfg_coef_slider], outputs=audio_output)

# Launch the interface.
demo.launch(show_error=True)