app.py

import gradio as gr
import utils
import clean
import transcribe

css = """
.cursive-text {
    font-family: 'Brush Script MT', cursive;
}
"""

with gr.Blocks(theme="base", css=css) as demo:
    gr.Markdown("<center><h1> 🔊 Transcription <span class='cursive-text'>Delight</span> </h1></center>")
    with gr.Row():
        with gr.Column():
            with gr.Row():
                source = gr.Radio(label="Source type", choices=[ ("YouTube URL", "youtube"), ("Audio", "audio"), ("Video", "video")], value="youtube")
                cleanup_options = gr.CheckboxGroup(label="Cleanup Transcript with LLM", choices=["Remove typos", "Separate into paragraphs"])
            @gr.render(inputs=source)
            def show_source(s):
                if s == "audio":
                    source_component = gr.Audio(type="filepath")
                elif s == "video":
                    source_component = gr.Video()
                else:
                    source_component = gr.Textbox(placeholder="https://www.youtube.com/watch?v=44vi31hehw4")
                    preview = gr.HTML(label="Video preview")
                    source_component.change(utils.convert_to_embed_url, source_component, preview)

                    transcribe_btn.click(
                        utils.generate_audio,
                        [source, source_component],
                        [download_audio],
                        show_progress="minimal"
                    ).success(
                        lambda : gr.Accordion(open=True),
                        None,
                        raw_accordion
                    ).then(
                        transcribe.transcribe,
                        [download_audio],
                        [preliminary_transcript],
                        show_progress="minimal"
                    ).success(
                        lambda : (gr.Accordion(open=False), gr.Accordion(open=True)),
                        None,
                        [raw_accordion, final_accordion]
                    ).then(
                        clean.clean_transcript,
                        [download_audio, cleanup_options, llm_prompt, preliminary_transcript],
                        [final_transcript, download_md],
                        show_progress="minimal"
                    )

        with gr.Column():
            with gr.Row():
                transcribe_btn = gr.Button("Transcribe 📜", variant="primary")
            with gr.Accordion("Raw transcript", open=False) as raw_accordion:
                preliminary_transcript = gr.Markdown("*Raw transcript will appear here*", show_label=False, height=400)
            with gr.Accordion("Final transcript", open=False) as final_accordion:
                final_transcript = gr.Markdown("*Final transcript will appear here*", height=400)

    source.change(utils.transcribe_button, source, transcribe_btn)

    with gr.Accordion("⚙️ Settings and Files", open=False) as settings_accordion:
        with gr.Row():
            with gr.Column():
                llm_prompt = gr.Textbox(label="LLM Prompt", visible=False, lines=3)
                cleanup_options.change(
                    utils.generate_prompt,
                    cleanup_options,
                    [llm_prompt, settings_accordion]
                )
                with gr.Row():
                    download_audio = gr.DownloadButton("Download .mp3 File 📥", interactive=False)
                    download_md = gr.DownloadButton("Download .md 📥", interactive=False)
    
demo.launch()