YoutubeTranscriber_streamlit.py

from youtube_transcript_api import YouTubeTranscriptApi
from urllib import parse
import streamlit as st
# import openai

list_languages = ["nl", "en", "de", "fr", "it", "sv", "da", "af", "sq", "am", "ar", "hy", "az", "eu", "be", "bn", "my",
                "bs", "bg", "ca", "ceb", "zh-Hant", "zh-Hans",
                "co", "eo", "et", "fil", "fi", "fr", "fy", "gl", "ka", "el", "gu", "ht", "ha", "haw", "iw",
                "hi", "hmn", "hu", "ga", "ig", "is", "id", "ja", "jv", "yi", "kn", "kk", "km", "rw", "ky", "ku", "ko",
                "hr", "lo", "la", "lv", "lt", "lb", "mk", "mg", "ml", "ms", "mt", "mi", "mr", "mn", "ne", "no", "ny",
                "or", "ug", "uk", "uz", "ps", "fa", "pl", "pt", "pa", "ro", "ru", "sm", "gd", "sr", "sn", "sd", "si", "sl",
                "sk", "su", "so", "es", "sw", "tg", "ta", "tt", "te", "th", "cs", "tk", "tr", "ur", "vi", "cy", "xh", "yo", "zu", "st"]

def transcribe_video(video_id, translate, language_from, translate_to):
    
    total_text, num_words = transcribe_really(video_id, translate, language_from, translate_to, list_languages)

    st.info(f"Number of words {num_words} / number of characters {len(total_text)} / number of tokens {int(len(total_text)/4)} (max.  4097) ")
    
    st.write(total_text)

    st.subheader("*To copy the text, roll over the mouse the box below and click the copy to clipboard icon at the right*")
    st.code (total_text)

    st.sidebar.markdown(
        '<br><br><a href="http://bark.phon.ioc.ee/punctuator" target="_blank">Go to Punctuator to punctuate the text</a>',
        unsafe_allow_html=True)

    st.sidebar.markdown(
        '<br><br><a href="https://beta.openai.com/playground/" target="_blank">Or go to Open AI playground to punctuate the text</a>',
        unsafe_allow_html=True)

def transcribe_really(video_id, translate, language_from, translate_to,list_languages):
    try:
        if translate == True:
            transcript_list = YouTubeTranscriptApi.list_transcripts(video_id)
            transcript = transcript_list.find_transcript([language_from])
            translated_transcript = transcript.translate(translate_to)
            transcript_fetched = (translated_transcript.fetch())
        else:
            transcript_fetched = YouTubeTranscriptApi.get_transcript(video_id,  languages=list_languages)
    except Exception as e:
        st.error(f"ERROR: {e}")
        return None, None
    
    total_text = ""
    for t in transcript_fetched:
        total_text += (t["text"])+ " "
    #st.write(total_text)
    num_words = len(total_text.split())
    return total_text,num_words

def do_punctuate(total_text):
    # VERY VERY SLOW !!!!!
    p = Punctuator('C:\\Users\\rcxsm\\AppData\\Local\\Programs\\Python\\Python38\\Lib\\site-packages\\punctuator\\Demo-Europarl-EN.pcl')
    punctuated_text = p.punctuate(total_text)
    print (punctuated_text)
    pc.copy(punctuated_text)
    print ("Text copied")

def get_video_id(url):
    if url != "":
        st.info (f"Transcribing {url }")
        url_parsed = parse.urlparse(url)
        video_id_ = parse.parse_qs(url_parsed.query)
        video_id__ = video_id_['v']
        for vx in video_id__:
            video_id = vx
        return video_id
    else:
        st.warning("Enter an URL to transcribe")

def main():
    st.sidebar.title ("Video transcriber")
  
    languages_full = ["Afrikaans", "Albanees", "Amhaars", "Arabisch", "Armeens", "Azerbeidzjaans", "Baskisch", "Belarussisch",
                "Bengaals", "Birmaans", "Bosnisch", "Bulgaars", "Catalaans", "Cebuano", "Chinees (Traditioneel)", "Chinees (Vereenvoudigd)",
                "Corsicaans", "Deens", "Duits", "Engels", "Esperanto", "Estisch", "Filipijns", "Fins", "Frans", "Fries", "Galicisch",
                "Georgisch", "Grieks", "Gujarati", "Haïtiaans Creools", "Hausa", "Hawaïaans", "Hebreeuws", "Hindi", "Hmong",
                "Hongaars", "Iers", "Igbo", "IJslands", "Indonesisch", "Italiaans", "Japans", "Javaans", "Jiddisch", "Kannada",
                "Kazachs", "Khmer", "Kinyarwanda", "Kirgizisch", "Koerdisch", "Koreaans", "Kroatisch", "Laotiaans", "Latijn",
                "Lets", "Litouws", "Luxemburgs", "Macedonisch", "Malagassisch", "Malayalam", "Maleis", "Maltees", "Maori",
                "Marathi", "Mongools", "Nederlands", "Nepalees", "Noors", "Nyanja", "Odia", "Oeigoers", "Oekraïens",
                "Oezbeeks", "Pasjtoe", "Perzisch", "Pools", "Portugees", "Punjabi", "Roemeens", "Russisch", "Samoaans",
                "Schots-Gaelisch", "Servisch", "Shona", "Sindhi", "Singalees", "Sloveens", "Slowaaks", "Soendanees",
                "Somalisch", "Spaans", "Swahili", "Tadzjieks", "Tamil", "Tataars", "Telugu", "Thai", "Tsjechisch",
                "Turkmeens", "Turks", "Urdu", "Vietnamees", "Welsh", "Xhosa", "Yoruba", "Zoeloe", "Zuid-Sotho", "Zweeds"]
    url = st.sidebar.text_input("Video to transcribe")
    translate = st.sidebar.selectbox("Translate", [True, False], index=1)
    if translate:
        language_from =  st.sidebar.selectbox("Translate from", list_languages, index=1)
        translate_to =  st.sidebar.selectbox("Translate to", list_languages, index=0)
    else:
        language_from, translate_to = None, None

    video_id = get_video_id(url)

    if video_id != None:
        transcribe_video(video_id, translate, language_from, translate_to)

    # summary = summarize_video(video_id)
    # st.write(summary)

def summarize_video(video_id):
    transcript = YouTubeTranscriptApi.get_transcript(video_id)
    text = "\n".join([t['text'] for t in transcript])
    try:
        resp = openai.ChatCompletion.create(
            model="gpt-4o-mini",
            messages=[{"role":"user", "content": f"Summarize:\n{text}"}]
        )
    except Exception as e:
        st.error(f"ERROR: {e}")
        return None
    return resp.choices[0].message.content


if __name__ == "__main__":
    main()