main.py

import streamlit as st
from youtube_transcript_api import YouTubeTranscriptApi
import os
import openai
from dotenv import load_dotenv

load_dotenv()

st.header("Get podcast notes using AI")

st.write('GitHub repo for this project: https://github.com/abekek/podcast_notes_chatgpt')

st.markdown("""---""")

yt_id = st.text_input('YouTube video ID')

bpm = st.slider('Choose # of bullet points/minute', 1, 5)

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

if st.button('Get notes'):
    result = YouTubeTranscriptApi.get_transcript(yt_id)

    curr_time = 0
    interval = 1
    text = []
    curr_text = ""
    for obj in result:
        curr_text += obj['text']
        curr_time = obj['start']
        if curr_time > interval * 600:
            text.append(curr_text)
            curr_text = ""
            interval += 1
        
    text.append(curr_text)

    notes = ""
    prompt = f"Based on the following YouTube transcript provide a {5*bpm} bullet point summary in complete sentences:\n"

    st.write('Number of note blocks: ' + str(len(text)))

    for i in range(len(text)):
        print('test')
        # res = api.send_message(prompt + text[i])['message'] + "\n"
        res = openai.Completion.create(
            model="text-davinci-003",
            prompt=prompt + text[i] + '\nSummary:\n',
            temperature=0.6,
            max_tokens=576,
            top_p=1,
            frequency_penalty=1,
            presence_penalty=1
        )
        res_text = res['choices'][0]['text']
        st.write('Block ' + str(i+1) + '/' + str(len(text)) + ':')
        st.write(res_text)
        notes += res_text