app.py

import os
import tempfile

import gradio as gr
from groq import Groq

from hf import hf_transcript, get_whisper_hf_client
from logs import configure_logging
from remote_llm import summarize
from transcribe import parse_audio, get_full_transcript, groq_transcript


def gradio_pipeline(video: str, whisper_inference, groq_api_token):
    groq_client = Groq(api_key=groq_api_token)
    hf_client = get_whisper_hf_client()
    print(repr(video))
    with tempfile.TemporaryDirectory() as tmpdirname:
        if video.endswith(".mp4"):
            parse_audio(video, os.path.join(tmpdirname, "audio.mp3"))
        elif video.endswith(".mp3"):
            os.replace(video, os.path.join(tmpdirname, "audio.mp3"))
        else:
            gr.Error("Unsupported file type")
        if whisper_inference == "hf":
            transcript = get_full_transcript(tmpdirname, hf_client, one_file_transcript_func=hf_transcript)
        elif whisper_inference == "groq":
            transcript = get_full_transcript(tmpdirname, groq_client, one_file_transcript_func=groq_transcript)
        return summarize(transcript, groq_client)


if __name__ == "__main__":
    configure_logging()
    demo = gr.Interface(
        fn=gradio_pipeline,
        title="Video summarizer",
        inputs=[
            gr.File(file_types=[".mp4", ".mp3"], label="Video or audio"),
            gr.Radio(choices=["groq", "hf"], value="hf", label="Whisper inference"),
            gr.Text(max_lines=1, type="password",
                    placeholder="Enter your groq API key",
                    label="groq API key")
        ],
        outputs=gr.Markdown(
            value="# Here will be the summary...",
            label="Summary",
            show_copy_button=True,
        ),
        examples=[
            ["examples/mipt.mp4"],
            ["examples/audio.mp3"],
        ],
        cache_examples=False,
        allow_flagging="never")
    demo.launch()