app.py

import gradio as gr
from pyannote.audio import Pipeline
from transformers import pipeline


asr = pipeline(
    "automatic-speech-recognition",
    model="facebook/wav2vec2-large-960h-lv60-self",
    feature_extractor="facebook/wav2vec2-large-960h-lv60-self",
    
)
speaker_segmentation = Pipeline.from_pretrained("pyannote/speaker-segmentation")

def segmentation(audio):
    speaker_output = speaker_segmentation(audio)
    text_output = asr(audio,return_timestamps="word")
    
    full_text = text_output['text'].lower()
    chunks = text_output['chunks']

    diarized_output = ""
    i = 0
    for turn, _, speaker in speaker_output.itertracks(yield_label=True):
        diarized = ""
        while i < len(chunks) and chunks[i]['timestamp'][1] <= turn.end:
            diarized += chunks[i]['text'].lower() + ' '
            i += 1
        
        if diarized != "":
            diarized_output += "{}: ''{}'' from {:.3f}-{:.3f}\n".format(speaker,diarized,turn.start,turn.end)
        
    return diarized_output, full_text

inputs = gr.inputs.Audio(source="upload", type="filepath", label="Upload your audio file here:")
outputs = [gr.outputs.Textbox(type="auto", label="Diarized Output"),
            gr.outputs.Textbox(type="auto",label="Full Text")]
examples = [["TestAudio1.wav"],]

app = gr.Interface(fn=segmentation,
                inputs=inputs,
                outputs=outputs,
                examples=examples,
                allow_flagging=False)
app.launch()