app.py

import gradio as gr
import torch
import torchaudio
import torchaudio.functional as AF
from hg_mms import Transcribe


def transcribe(audio_file, lang_id: str):
    print(f"audio_file={audio_file}")
    print(lang_id)
    freq = 16000
    # Return the transcript.
    transcript = ""
    # load the auido file to tensor
    waveform, orig_freq = torchaudio.load(audio_file.name)
    # resample audio to 16Khz
    if orig_freq != freq:
        waveform = AF.resample(waveform, orig_freq, freq)

    return transcriber(waveform, lang_id), audio_file.name

if __name__ == "__main__":
    transcriber = Transcribe()
    inputs = [gr.File(), gr.Dropdown(choices=["amh", "orm", "som"])]
    outputs = [
        gr.Textbox(label="Transcript"),
        gr.Audio(label="Audio", type="filepath"),
    ]

    app = gr.Interface(transcribe, inputs=inputs, outputs=outputs)
    app.launch()