app.py

import gradio as gr
from transformers import pipeline
import librosa

pipe = pipeline("automatic-speech-recognition", model="model")


def transcribe(audiofile):
    audio, sr = librosa.load(audiofile, sr=None)
    if sr != 16_000:
        audio = librosa.resample(audio, orig_sr=sr, target_sr=16_000)

    text = pipe(audio, chunk_length_s=25, stride_length_s=(5, 5))['text']

    return text


demo = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(type='filepath'),
    outputs=gr.Textbox(show_copy_button=True)
)

if __name__ == '__main__':
    demo.launch()