app.py

import gradio as gr
import torch
from transformers import pipeline
from timestamp import format_timestamp

MODEL_NAME = "openai/whisper-medium"
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(file, task, return_timestamps):
    outputs = pipe(file, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)
    text = outputs["text"]
    timestamps = outputs["chunks"]
    
    if return_timestamps==True:
      timestamps = [f"[{format_timestamp(chunk['timestamp'][0])} -> {format_timestamp(chunk['timestamp'][1])}] {chunk['text']}" for chunk in timestamps]
      
    else:
      timestamps = [f"{chunk['text']}" for chunk in timestamps]
      
    text = "\n".join(str(feature) for feature in timestamps)    
    return text


file_transcribe = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.inputs.Audio(source="upload", optional=True, label="Audio file", type="filepath"),
        gr.inputs.Radio(["transcribe"], label="Task", default="transcribe"),
        gr.inputs.Checkbox(default=False, label="Return timestamps"),
    ],
    outputs="text",
    layout="horizontal",
    theme="huggingface",
    title="Whisper Demo: Transcribe Audio",
    description=(
        "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the"
        f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
        " of arbitrary length."
    ),
    
    cache_examples=True,
    allow_flagging="never",
)

file_transcribe.launch(enable_queue=True, debug = True)