app.py

import gradio as gr
from transformers import pipeline
import torch

MODEL_NAME = "openai/whisper-large-v3"
BATCH_SIZE = 8

device = 0 if torch.cuda.is_available() else "cpu"

pipe = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
)

def transcribe(audio):
  print(audio)
  result = pipe(audio, batch_size=BATCH_SIZE)["text"]
  print(result)

  return result

demo = gr.Blocks()

app = gr.Interface(fn=transcribe, inputs=gr.inputs.Audio(source="microphone", type="filepath"), outputs="textbox")
with demo:
  gr.TabbedInterface([app], "Mic")

demo.launch()