app.py

import os
import openai 
import gradio as gr 
from youtube_transcript_api import YouTubeTranscriptApi
from urllib.parse import urlparse, parse_qs
from requests.structures import CaseInsensitiveDict

openai.api_key = os.getenv("OPENAI_API_KEY")

google_analtycs="""
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-S9JEXRFQJF"></script>
<script>
  window.dataLayer = window.dataLayer || [];
  function gtag(){dataLayer.push(arguments);}
  gtag('js', new Date());

  gtag('config', 'G-S9JEXRFQJF');
</script>
"""


def Prompt_T(context, lang):

  prompt  = """I want you to act as a content writer who is working with youtube video transcript. Summarise the following text in 70 words:
  =========
  """+ context +"""
  =========
  Answer:"""

  if (lang=="ru"):prompt  = """Я хочу, чтобы вы выступили в роли автора контента, который взят с  транскрипт youtube видео, его нужно преобразовать  в читаемый вид. Резюмируйте следующий текст в 50 слов:
  =========
  """+ context +"""
  =========
  Ответ:"""

  if (lang=="uk"):prompt  = """Я хочу, щоб ви виступили в ролі автора контенту, який узятий з транскрипту youtube відео,  його треба перетворити у читабельний вигляд. Резюмуйте наступний текст у 50 слів: 
  =========
  """+ context +"""
  =========
  Ответ:"""

  return prompt


def split_string(string, chunk_size):
    return [string[i:i+chunk_size] for i in range(0, len(string), chunk_size)]


def gpt_api (input_text):
  completion = openai.ChatCompletion.create(
    model="gpt-3.5-turbo",
    messages=[  {"role": "system", "content": input_text} ]
    )
  response = completion.choices[0].message.content
  return response

def generate(video_url, request: gr.Request):

    #Если есть get переменная v, создаем video_url
    try:
      headers =request.headers
      url = headers.get('referer')
      parsed_url = urlparse(url)
      query_params = parse_qs(parsed_url.query)
      my_dict = query_params
    except KeyError as e:
      print("Что то случилось в header", e)
      
    try:
      my_v = my_dict['v'][0]
      video_url ="https://youtube.com/watch?v="+my_v
    except KeyError:
      my_v = ""

    #Если две переменные пустые, то показываем базовую страницу с рекламой
    if (video_url =="") and (my_v == ""):  
      html_embed='<div><br> An easy way to get video descriptions If you are on YouTube itself, simply add "zxc" in front of YouTube to the videos address.</div>'
      summarize=""
      return summarize, html_embed
    
    #похоже ли video_url на номальну ссылку
    if "youtube.com/watch?v=" in video_url: x=111
    else: return "Неверный URL", "Ошибка"
 
    #Пробуем извлеч video_id пока на английском
    video_id = video_url[-11:]

    try:
      transcript_list = YouTubeTranscriptApi.list_transcripts(video_id)
    except Exception as e:
      return "No access for transcript ", "Error transcript_list" 
    
    # iterate over all available transcripts
    lang_video="en"
    for transcript in transcript_list:
      if (transcript.is_generated == True): lang_video = transcript.language_code
      print ("transcript.language_code=", transcript.language_code)
      
    try:
      t = YouTubeTranscriptApi.get_transcript(video_id,languages=[lang_video])
      # do something with the transcript
    except Exception as e:
      return "No access for transcript", "Error transcript"
      
    finalString = ""
    for item in t:
        text = item['text']
        finalString += text + " "

  
    print("Transcript:",finalString)
    print("Transcript lenght:",len(finalString))
    print ("===============================================")
    input_string = finalString


    chunk_size = 12000
    if (lang_video=="ru"): chunk_size = 5000
    if (lang_video=="uk"): chunk_size = 5000

    result_list = split_string(input_string, chunk_size)
    final_answer_gpt=""
    count= 0

    print("++++++++++++++++++++++++++++++++++++++")
    for item in result_list:
        print(item)


    for item in result_list:
      count = count +1
      context = item
      input_gpt = Prompt_T(context,lang_video)
      final_answer_gpt = final_answer_gpt +"<p>" + gpt_api (input_gpt)+"</p>"

      
    html_embed='<iframe width="450" height="158" src="https://www.youtube.com/embed/'+ video_id +'" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe>'
    html_content="<h6>"+"<br>"+final_answer_gpt+"</h6>"
    return html_content, html_embed

title = "YouTube Summorize (en,ua,ru)"
css="""
footer {visibility: hidden}
.gradio-container {padding-top: 100px}
"""
with gr.Blocks(css=css, title=title) as demo:
    gr.HTML("<h1>A simple way to summarise and translate the YouTube video in 22 languages</h1>"+google_analtycs)
    with gr.Row():   
        with gr.Column():
          input_d = gr.Textbox(label="YouTube video URL", placeholder="https://www.youtube.com/watch?v=XXXXXXXX", value="")
          greet_btn = gr.Button("Summarise")
        dt_2 = gr.outputs.HTML() 
    dt_1 = gr.outputs.HTML()
    dt =[dt_1, dt_2]
    greet_btn.click(generate, inputs=input_d, outputs=dt)
    demo.load(generate, inputs=input_d, outputs=dt)

demo.launch(share=False, debug=True )