app.py

import os
from math import floor
from typing import Optional

import spaces
import torch
import gradio as gr
from transformers import pipeline
from transformers.pipelines.audio_utils import ffmpeg_read

# config
model_name = "kotoba-tech/kotoba-whisper-v2.2"
example_file = "sample_diarization_japanese.mp3"

# device setting
if torch.cuda.is_available():
    torch_dtype = torch.bfloat16
    device = "cuda"
    model_kwargs = {'attn_implementation': 'sdpa'}
else:
    torch_dtype = torch.float32
    device = "cpu"
    model_kwargs = {}

# define the pipeline
pipe = pipeline(
    model=model_name,
    chunk_length_s=15,
    batch_size=16,
    torch_dtype=torch_dtype,
    device=device,
    model_kwargs=model_kwargs,
    trust_remote_code=True
)
sampling_rate = pipe.feature_extractor.sampling_rate


def format_time(start: Optional[float], end: Optional[float]):

    def _format_time(seconds: Optional[float]):
        if seconds is None:
            return "[no timestamp available]"
        minutes = floor(seconds / 60)
        hours = floor(seconds / 3600)
        seconds = seconds - hours * 3600 - minutes * 60
        m_seconds = floor(round(seconds - floor(seconds), 1) * 10)
        seconds = floor(seconds)
        return f'{minutes:02}:{seconds:02}.{m_seconds:01}'

    return f"[{_format_time(start)} -> {_format_time(end)}]:"


@spaces.GPU
def get_prediction(inputs):
    return pipe(inputs, generate_kwargs={"language": "ja", "task": "transcribe"})


def transcribe(inputs: str):
    if inputs is None:
        raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
    with open(inputs, "rb") as f:
        inputs = f.read()
    prediction = get_prediction({"array": ffmpeg_read(inputs, sampling_rate), "sampling_rate": sampling_rate})
    output = ""
    for n, s in enumerate(prediction["speakers"]):
        text_timestamped = "\n".join([f"- **{format_time(*c['timestamp'])}** {c['text']}" for c in prediction[f"chunks/{s}"]])
        output += f'### Speaker {n+1} \n{text_timestamped}\n'
    return output


description = (f"Transcribe and diarize long-form microphone or audio inputs with the click of a button! Demo uses "
               f"Kotoba-Whisper [{model_name}](https://huggingface.co/{model_name}).")
title = f"Audio Transcription and Diarization with {os.path.basename(model_name)}"
shared_config = {"fn": transcribe, "title": title, "description": description, "allow_flagging": "never", "examples": [example_file]}
o_upload = gr.Markdown()
o_mic = gr.Markdown()
i_upload = gr.Interface(
    inputs=[gr.Audio(sources="upload", type="filepath", label="Audio file")], outputs=gr.Markdown(), **shared_config
)
i_mic = gr.Interface(
    inputs=[gr.Audio(sources="microphone", type="filepath", label="Microphone input")], outputs=gr.Markdown(), **shared_config
)
with gr.Blocks() as demo:
    gr.TabbedInterface([i_upload, i_mic], ["Audio file", "Microphone"])
demo.queue(api_open=False, default_concurrency_limit=40).launch(show_api=False, show_error=True)