app.py

import torch
import gradio as gr
from transformers import pipeline

model_id = "Sandiago21/whisper-large-v2-italian"  # update with your model id
pipe = pipeline("automatic-speech-recognition", model=model_id)


title = "Automatic Speech Recognition (ASR)"
description = """
Demo for automatic speech recognition in Italian. Demo uses [Sandiago21/whisper-large-v2-italian](https://huggingface.co/Sandiago21/whisper-large-v2-italian) checkpoint, which is based on OpenAI's
[Whisper](https://huggingface.co/openai/whisper-large-v2) model and is fine-tuned in Italian Audio dataset
![Automatic Speech Recognition (ASR)"](https://datasets-server.huggingface.co/assets/huggingface-course/audio-course-images/--/huggingface-course--audio-course-images/train/2/image/image.png "Diagram of Automatic Speech Recognition (ASR)")
"""

def transcribe_speech(filepath):
    output = pipe(
        filepath,
        max_new_tokens=256,
        generate_kwargs={
            "task": "transcribe",
            "language": "italian",
        },  # update with the language you've fine-tuned on
        chunk_length_s=30,
        batch_size=8,
    )
    return output["text"]

demo = gr.Blocks()

mic_transcribe = gr.Interface(
    fn=transcribe_speech,
    inputs=gr.Audio(source="microphone", type="filepath"),
    outputs=gr.outputs.Textbox(),
    tilte=title,
    description=description,
)

file_transcribe = gr.Interface(
    fn=transcribe_speech,
    inputs=gr.Audio(source="upload", type="filepath"),
    outputs=gr.outputs.Textbox(),
    examples=[["./example.wav"]],
    tilte=title,
    description=description,
)

with demo:
    gr.TabbedInterface(
        [mic_transcribe, file_transcribe],
        ["Transcribe Microphone", "Transcribe Audio File"],
    ),

demo.launch()