app.py

import gradio as gr
import numpy as np
from vad_utils import get_speech_probs, make_visualization, probs2speech_timestamps, read_audio
import torch
import pandas as pd
import gdown

def process_audio(audio_input, window_size_samples):
    wav = read_audio(audio_input, sampling_rate=16_000)
    audio_length_samples = len(wav)
    probs = get_speech_probs(wav, window_size_samples=window_size_samples, sampling_rate=16_000)
    return make_visualization(probs, 512 / 16_000), probs, audio_length_samples

def process_parameters(probs, audio_length_samples, threshold, min_speech_duration_ms, min_silence_duration_ms, window_size_samples, speech_pad_ms):
    min_speech_duration_ms *= 1000
    min_silence_duration_ms *= 1000
    timestamps = probs2speech_timestamps(probs, audio_length_samples, 
                                         threshold = threshold,
                                         min_speech_duration_ms = min_speech_duration_ms, 
                                         min_silence_duration_ms=min_silence_duration_ms, 
                                         window_size_samples=window_size_samples, 
                                         speech_pad_ms=speech_pad_ms,
                                         return_seconds=True,
                                         rounding=3)
    print(timestamps)
    df = pd.DataFrame(timestamps)
    df["note"] = ""
    df.to_csv("timestamps.txt", sep = '\t', header=False, index=False)
    return "timestamps.txt", df

def download_gdrive(id):
    output_file = "audio.wav"  # Replace "data_file.ext" with the desired output filename and extension

    gdown.download(f"https://drive.google.com/uc?id={id}", output_file)
    return output_file

def main(): 
    with gr.Blocks() as demo:
        probs = gr.State()
        audio_length_samples = gr.State()
        with gr.Row():
            info = """Input the Google Drive file id from the shared link.
            It comes after https://drive.google.com/file/d/ <id here.
            For example the link https://drive.google.com/file/d/15C6aHry8sJr43r0EYPPrIlPjMWp6SDb8/view?usp=drive_link has id 15C6aHry8sJr43r0EYPPrIlPjMWp6SDb8"""
            gdrive_str = gr.Text(label="File ID", info = info)
            download_button = gr.Button("Download Audio")

        with gr.Row():
            audio_input = gr.Audio(type="filepath")
            with gr.Column():
                md = gr.Markdown("[Parameter Documentation](https://github.com/snakers4/silero-vad/blob/master/utils_vad.py#L198)")
                window_size_samples = gr.Dropdown(label="Window Size (samples)", choices=[512, 1024, 1536], value=512)
                button1 = gr.Button("Compute Speech Probabilities")
            figure = gr.Plot()

        download_button.click(download_gdrive, inputs=[gdrive_str], outputs=audio_input)

        button1.click(process_audio, inputs=[audio_input, window_size_samples], outputs=[figure, probs, audio_length_samples])

        with gr.Row():
            threshold = gr.Number(label="Threshold", value=0.6, minimum=0.0, maximum=1.0)
            min_speech_duration_ms = gr.Number(label="Mininmum Speech Duration (s)", value=10.5)
            min_silence_duration_ms = gr.Number(label="Minimum Silence Duration (s)", value=5.5)
            speech_pad_ms = gr.Number(label="Speech Pad (ms)", value=30)
            button2 = gr.Button("Compute Speech Timestamps")
            output_file = gr.File()
        with gr.Row():
            output_df = gr.DataFrame()

        button2.click(process_parameters, inputs=[probs, audio_length_samples, threshold, min_speech_duration_ms, min_silence_duration_ms, window_size_samples, speech_pad_ms],
                      outputs=[output_file, output_df])

    demo.launch()

if __name__ == "__main__":
    main()