app.py

from transformers import pipeline, WhisperProcessor
import gradio as gr
import time

checkpoint_name = "openai/whisper-medium"
p = pipeline("automatic-speech-recognition", model=checkpoint_name)
processor = WhisperProcessor.from_pretrained(checkpoint_name)


def transcribe(audio, language, task):
    forced_decoder_ids = processor.get_decoder_prompt_ids(language=language, task=task)
    mic_audio_transcription = p(audio, generate_kwargs={"forced_decoder_ids": forced_decoder_ids})["text"] if audio else ""
    return mic_audio_transcription

gr.Interface(
    fn=transcribe,
    inputs=[gr.Audio(source="microphone", type="filepath"), 
            gr.Radio(label="language", choices=['english', 'german', 'spanish', 'italian', 'french', 'polish'], value='english'), 
            gr.Radio(label="task", choices=['transcribe', 'translate'], value='transcribe')],
    outputs=["text"]).launch()