app.py

import gradio as gr
import os
import torch
import io
import wavio
import numpy as np
from pyannote.audio import Pipeline
from pyannote.audio import Audio
from pyannote.core import Segment

pipeline = Pipeline.from_pretrained(
  "pyannote/speaker-diarization-3.1",
  use_auth_token=os.environ['api'])

def process_audio(audio):
    # Extract the audio data and sample rate from the tuple
    audio_data, sample_rate = audio
    
    # Ensure the audio data is in the correct format
    audio_data = np.int16(audio_data / np.max(np.abs(audio_data)) * 32767)
    
    # Convert audio_data to a numpy array
    audio_data = np.array(audio_data)
    
    # Save the uploaded audio file to a temporary location
    wavio.write("temp.wav", audio_data[:, np.newaxis], sample_rate, sampwidth=2)
    
    # Use the diarization pipeline to process the audio
    diarization = pipeline("temp.wav")
    
    # Remove the temporary file
    os.remove("temp.wav")
    
    # Return the diarization output
    return diarization

with gr.Blocks() as demo:
    audio_input = gr.Audio(label="Upload Audio")  
    process_button = gr.Button("Process")
    diarization_output = gr.JSON(label="Diarization Output")

    process_button.click(fn=process_audio, inputs=audio_input, outputs=diarization_output)

demo.launch()