app.py

from pyharp import ModelCard, build_endpoint, save_and_return_filepath
from audiotools import AudioSignal
from audioldm import build_model, style_transfer
import gradio as gr
import soundfile as sf

audioldm = build_model(model_name="audioldm-m-full")

def process_fn(input_audio_path, prompt, seed, guidance_scale, num_inference_steps, num_candidates, audio_length_in_s, transfer_strength):
    waveform = style_transfer(
        audioldm,
        prompt,
        input_audio_path,
        transfer_strength,
        int(seed),
        duration = audio_length_in_s,
        guidance_scale = guidance_scale,
        ddim_steps = int(num_inference_steps),
        batchsize = int(num_candidates),
        config=None,
    )
    waveform = waveform[:,None,:]
    sf.write("./output.wav", waveform[0, 0], samplerate=16000)
    
    #save_wave(waveform, "./", name="output.wav")  #broken, always appends _0.wav
    return "./output.wav"


card = ModelCard(
    name='AudioLDM Variations',
    description='AudioLDM Variation Generator, operates on region selected in track.',
    author='Team Audio',
    tags=['AudioLDM', 'Variations', 'audio-to-audio']
)

with gr.Blocks() as webapp:
    # Define your Gradio interface
    inputs = [
        gr.Audio(
            label="Audio Input", 
            type="filepath"
        ), 
        gr.Text(
            label="Prompt", 
            interactive=True
        ),
        gr.Slider(
            label="seed",
            minimum="0",
            maximum="65535",
            value="43534",
            step="1"
        ),
        gr.Slider(
            minimum=0, maximum=10, 
            step=0.1, value=2.5, 
            label="Guidance Scale"
        ),
        gr.Slider(
            minimum=1, maximum=500, 
            step=1, value=200, 
            label="Inference Steps"
        ),
        gr.Slider(
            minimum=1, maximum=10, 
            step=1, value=1, 
            label="Candidates"
        ),
        gr.Slider(
            minimum=2.5, maximum=10.0, 
            step=2.5, value=5, 
            label="Duration"
        ),
        gr.Slider(
            minimum=0, maximum=1.0, 
            value=0.5, 
            label="Transfer Strength"
        ),
    ]

    output = gr.Audio(label="Audio Output", type="filepath")

    ctrls_data, ctrls_button, process_button, cancel_button = build_endpoint(inputs, output, process_fn, card)

# queue the webapp: https://www.gradio.app/guides/setting-up-a-demo-for-maximum-performance
#webapp.queue()
webapp.launch(share=True)