app.py

import torchaudio
from torchaudio.functional import resample
import os
from pyannote.audio import Pipeline
os.environ["PYANNOTE_SKIP_DEPENDENCY_CHECK"] = "1"

def process(input_file):
    pipeline = Pipeline.from_pretrained("shethjenil/speaker-diarization-community-1")
    audio, sr = torchaudio.load(input_file)
    target_sr = 16000
    if sr != target_sr:
        audio = resample(audio, sr, target_sr)
    if audio.shape[0] > 1:
        audio = audio.mean(dim=0, keepdim=True)
    output = pipeline({"waveform":audio,"sample_rate":target_sr})
    return {
      "diarization":[[i['start'],i['end'],int(i['speaker'].lstrip("SPEAKER_"))] for i in output.serialize()['diarization']],
      "exclusive_diarization":[[i['start'],i['end'],int(i['speaker'].lstrip("SPEAKER_"))] for i in output.serialize()['exclusive_diarization']],
      "embedding":output.speaker_embeddings.tolist()
    }

import gradio as gr
gr.Interface(process, inputs=gr.Audio(type="filepath"), outputs=gr.JSON()).launch()