app.py

import json
import gradio as gr
import numpy as np
import matplotlib.pyplot as plt
from manipulate_model.utils import get_config_and_model, infere


manpulate_config, manipulate_model = get_config_and_model()
manipulate_model.eval()

def process(filepath):
    global manipulate_model
    global manpulate_config
    out = infere(manipulate_model, filepath, manpulate_config)
    out = out.tolist()

    #plt.clf()
    #plt.figure()
    #plt.plot(out)
    #out_masked = np.ma.masked_less_equal(out, 0.4)
    #plt.plot(out_masked, 'r', linewidth=2)
    #return str(out), plt
    output_json = {}
    output_json["decision_scores"] = str(out)
    response_text = json.dumps(output_json, indent=4)
    return response_text

demo = gr.Blocks()
file_proc = gr.Interface(
    fn=process,
    inputs=[
        #gr.Audio(sources=["microphone", "upload"], type="filepath", show_download_button=True, label="Speech file (<30s)", max_length=30),
        gr.Audio(sources=["upload"], label="Speech file (<30s)", type="filepath")
    ],
    outputs="text", #gr.Plot(label="Frame wise prediction")
    title="Find the manipulation: Analyze 'Real' or 'Manipulated' audio.",
    description=(
        "Analyze, detect and localize manipulation in an audio with a click of a button. Upload a .wav or .flac file."
    ),
    examples=[
        ["./samples/fake_audio.wav"],
        ["./samples/real_audio.wav"]
    ],
    cache_examples=False,
    allow_flagging="never",
)

with demo:
    gr.TabbedInterface([file_proc], ["Find Audio Manipulation"])
demo.queue(max_size=10)
demo.launch(share=True)