app.py

import os
import json
import random
import pymongo
import requests
import gradio as gr
from collections import defaultdict
from uuid import uuid4


DB_CONN = pymongo.MongoClient(os.environ.get("DB_URL"))
line_details = {l["source_id"]:l for l in list(DB_CONN['vo_data_dump']['qa_video_dump'].find()) if l.get("done") is not True}

videos_list = defaultdict(list)
for line_detail in line_details.values():
    videos_list[line_detail["video_id"]].append(line_detail)


# Functions
def render_video_details(video_id):
    video_detail = videos_list[video_id]
    video_title = video_detail[0]["video_title"]
    source_language = video_detail[0]["source_language"]
    target_language = video_detail[0]["target_language"]
    video_duration = video_detail[0]["video_duration"]
    line_ids = [v["source_id"] for v in video_detail if v.get("done") is not True]
    video_link = video_detail[0]["video_link"]
    html = f"""<iframe width="560" height="315" src="{video_link.replace('watch?v=', 'embed/')}" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe>"""
    if video_link.startswith("gs://"):
        video_html_embed, video_streamable_url = None, requests.post(os.environ["CF_URL"], params={"url": video_link}).text
    else:
        video_html_embed, video_streamable_url = html, None
    return video_title, source_language, target_language, video_duration, gr.Dropdown.update(choices=line_ids, value=line_ids[0]), video_html_embed, video_streamable_url

def render_line_details(line_id):
    line_detail = line_details[line_id]
    source_transcript = line_detail["source_transcript"]
    translation = line_detail["translator_translated_text"]
    response = requests.post(os.environ["CF_URL"], params={"url": line_detail["source_audio_url"]})
    signed_uri = response.text
    audio_response = requests.get(signed_uri).content
    return audio_response, source_transcript, translation

def dump_data_db(line_id, video_type, video_subject, video_topics, video_tone, gender, demography, total_time, source_language, target_language, comments, accent, source_transcript, translation, speaker_gender):
    line_detail = line_details[line_id]
    video_id = line_detail["video_id"]
    line_data = {
        "line_id": line_id,
        "video_id": video_id,
        "src_text": source_transcript,
        "intermediate_text": line_detail["english_transcript"],
        "tgt_text": translation,
        "accent": accent,
        "speaking_rate": round(len(translation.split()) / line_detail["line_duration"], 3),
        "src_time": line_detail["line_duration"],
        "gender": speaker_gender,
        "version": "v1",
        "done": True
    }
    video_data = {
        "video_id": video_id,
        "type": video_type,
        "src_lang": source_language,
        "tgt_lang": target_language,
        "subject": video_subject,
        "topics": video_topics,
        "tone": video_tone,
        "gender": gender,
        "demography": demography,
        "total_time": float(total_time),
        "comments": comments
    }
    video_data_collection = DB_CONN['vo_data_dump']['video_data']
    video_data_collection.update_one({'video_id': video_id}, {"$set": video_data}, upsert=True)
    
    line_data_collection = DB_CONN['vo_data_dump']['line_data']
    line_data_collection.update_one({'line_id': line_id}, {"$set": line_data}, upsert=True)

    DB_CONN['vo_data_dump']['qa_video_dump'].update_one({'source_id': line_id}, {"$set": {"done": True}}, upsert=True)

    # Delete video
    i = -1
    for i in range(len(videos_list[video_id])):
        if videos_list[video_id][i]["source_id"] == line_id:
            del videos_list[video_id][i]
            break
    if len(videos_list[video_id]):
        source_ids = [v["source_id"] for v in videos_list[video_id]]
        return gr.Dropdown.update(choices=source_ids, value=source_ids[i]), video_id
    else:
        del videos_list[video_id]
        new_video_id = random.choice(list(videos_list.values()))[0]['video_id']
        source_ids = [v["source_id"] for v in videos_list[new_video_id] if v.get('done') is not True]
        return gr.Dropdown.update(choices=source_ids, value=source_ids[0]), gr.Dropdown.update(choices=list(videos_list.keys()), value=new_video_id)


# UI
with gr.Blocks() as demo:
    gr.Markdown("## Data Aggregation")
    with gr.Row():
        video_id_dropdown = gr.Dropdown(list(videos_list.keys()), label="Video IDs")
        video_title_text = gr.Textbox(label="Video Title", interactive=False)
    with gr.Row():
        video_html = gr.HTML(label="Video")
        video_playable = gr.Video(label="Video")
    gr.Markdown("""Please ensure to fill these sections and do not leave them empty. An example of how to update `Video Type`, `Video Subject` and `Video Topics`:
- If the video is of a `Biology lecture`, **Video Type**: _learning_, **Video Subject**: _science_, **Video Topics**: _biology_
- If the video is of `Cooking`, **Video Type**: _cooking_, **Video Subject**: _cooking_, **Video Topics**: _indian recipe_""")
    with gr.Row():
        video_type_dropdown = gr.Dropdown(["learning", "audiobook", "podcast", "vlog", "news", "cooking", "review"], label="Video Type", value="learning", allow_custom_value=True)
        video_subject_dropdown = gr.Dropdown(["tech", "science", "lifestyle", "cooking", "travel", "finance", "politics"], label="Video Subject", value="tech", allow_custom_value=True)
        video_topics_text = gr.Textbox(label="Video Topics")
    with gr.Row():
        video_tone_dropdown = gr.Dropdown(["casual", "semi-formal", "formal"], label="Video Tone", value="semi-formal", info="Tone used in the video.")
        gender_dropdown = gr.Dropdown(["male", "female", "non-binary"], label="Gender", value="female", info="Gender used by the speaker in the video as whole.")
        demography_text = gr.Textbox(label="Demography", info="Demography like normal, old, young, toddler, etc.")
    with gr.Row():
        source_language_text = gr.Textbox(label="Source Language", interactive=False, visible=False)
        target_language_text = gr.Textbox(label="Target Language", interactive=False, visible=False)
    with gr.Row():
        total_time_text = gr.Textbox(label="Video Duration", interactive=False)
        comments_text = gr.Textbox(label="Comments", info="Any extra comments regarding the nature of video.")
    gr.Markdown("Video Line Information")
    line_id_dropdown = gr.Dropdown([], label="Line IDs")
    with gr.Row():
        audio_item = gr.Audio(label="Source Audio", type="filepath")
        accent_text = gr.Textbox(label="Accent", value="indian", info="Accent of the person speaking. For example, indian, american, british.")
        speaker_gender_dropdown = gr.Dropdown(["female", "male"], label="Audio Gender", value="female", info="Gender of the speaker")
    with gr.Row():
        source_transcript_text = gr.Textbox(label="Source Transcript", info="Transcription of the above rendered audio. Please ensure correct punctuations based on the audio.")
        translation_text = gr.Textbox(label="Translation", interactive=False, visible=False)
    update_button = gr.Button("Update")

    # Actions
    video_id_dropdown.change(render_video_details, video_id_dropdown, [video_title_text, source_language_text, target_language_text, total_time_text, line_id_dropdown, video_html, video_playable])
    line_id_dropdown.change(render_line_details, line_id_dropdown, [audio_item, source_transcript_text, translation_text])
    update_button.click(dump_data_db, [line_id_dropdown, video_type_dropdown, video_subject_dropdown, video_topics_text, video_tone_dropdown, gender_dropdown, demography_text, total_time_text, source_language_text, target_language_text, comments_text, accent_text, source_transcript_text, translation_text, speaker_gender_dropdown], [line_id_dropdown, video_id_dropdown])


if __name__=="__main__":
    demo.queue().launch(auth=(os.environ.get("USERNAME"), os.environ.get("PASSWORD")))