import csv import pytube from pytube import YouTube import moviepy.editor as mp from pydub import AudioSegment from urllib.parse import urlparse import os import whisper import openai import json import ffmpeg import yt_dlp model = whisper.load_model("large") def Analyze(video_link): video_url = f"{video_link}" # Your existing code for processing the video and analyzing with ChatGPT # Extracting details of the Video print("Extracting Video Details"'\n') try: video = YouTube(video_url) video_channel = video.author video_title = video.title video_description = video.description video_url = video.watch_url print(f"Channel Name: {video_channel}") print(f"Video Tile: {video_title}") print(f"Video URL: {video_url}") print(f"Video Description: {video_description}") except Exception as e: print(f"An error occured: {str(e)}") #--- try: # Determine if it's for YouTube or YouTube Shorts and save the audio \ # file according to its video ID parsed_url = urlparse(video_url) if 'youtube.com' in parsed_url.netloc: if '/shorts/' in parsed_url.path: # It's a YouTube Shorts URL video_id = parsed_url.path.split('/shorts/')[-1] else: # It's a regular YouTube video URL query_params = parsed_url.query.split('&') video_id = None for param in query_params: if param.startswith('v='): video_id = param[2:] break if video_id is None: print("Invalid YouTube video URL. No video ID found.") else: print("Invalid URL. Please provide a valid YouTube URL.") #--- # Downloading audio and converting to .wav format print("Downloading Audio") if video_id: # Replace 'VIDEO_URL' with the actual URL of the video you want to download # Options to download only audio in the best quality available options = { 'format': 'bestaudio/best', 'postprocessors': [{ 'key': 'FFmpegExtractAudio', 'preferredcodec': 'mp3', 'preferredquality': 'best', }], 'outtmpl': '%(id)s.%(ext)s', 'quiet': True, # Set quiet to True to suppress verbose output } # Create yt-dlp object with specified options ydl = yt_dlp.YoutubeDL(options) # Download the video ydl.download([video_url]) print("AUDIO DOWNLOADED") savepath_audio = f'{video_id}.mp3' print(savepath_audio) print('\n'"Audio path: ", savepath_audio) #Whisper Transcription print('\n'"Transcribing Audio, Please wait!"'\n') result = model.transcribe(savepath_audio, verbose=True, task='translate') os.remove(savepath_audio) combined_data = "" for segment in result['segments']: start_time = "[{:02d}:{:06.3f}".format(int(segment['start']) // 60, segment['start'] % 60) end_time = "{:02d}:{:06.3f}]".format(int(segment['end']) // 60, segment['end'] % 60) timestamp = start_time + ' --> ' + end_time text = segment['text'] # Combining transcript into a single string with newlines combined_data += timestamp + ' ' + text + '\n' #---- #Using ChatGPT for Analysis openai.api_key = os.environ.get('API_KEY') def callChatGPT(input_text): #input_prompt = "I have receive the message following message: " + '"' + str(input_text) + '"' + "." input_prompt = """you will be provided a transcript of a video with timestamp. perform the following actions keeping \ youtube commmunity guidelines, giving importance to freedom of speech and differentiating between fact and opinion,\ and the idea that people are free to believe what they want to believe and say what they want about other people with\ different viewpoints \ (as long as the latter do not belong to 'protected group' category i.e., video should not hateful towards age, caste, \ disability, ethnicity, gender identity and expression, nationality, race, immigration status, religion,\ sex/gender, sexual orientation, victims of a major violent event and their kin, and veteran status; if it is not \ against any protected group, then \ flag ONLY if the language is extremely hateful - ignore minor everyday foul deragatory language) in mind: 1 - summarize the text delimited by triple backticks with 1-2 sentence \ 2 - provide the overall sentiment of the text \ 3 - differentiate between fact or opinion - if it is an opinion, flag the text as not hateful by default unless there are explicit/hateful words \ 4 - check for any hate in the text. if it does then just say yes else say no \ 5 - if the text contains hate then mention all the timestamps and it's texts where it contains hate \ 6 - classify the risk/degree of hate into high, medium, or low \ 7 - if flagged as hate, explain why 8 - output a json object which contains the following keys: \ summary, sentiment, presence_of_hate, hate_degree, part_which_contains_hate, fact_or_opinion, explanation""" + '```' + str(input_text) + '```' + "." response = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ {"role": "user", "content": input_prompt} ], temperature=0, ) return response input_text = (f"""{combined_data}""") print('\n'"ANALYZING..."'\n') response_AI = callChatGPT(input_text) # Extracting values from the response response_data = response_AI['choices'][0]['message']['content'] response_json = json.loads(response_data) summary = response_json.get('summary', '') sentiment = response_json.get('sentiment', '') presence_of_hate = response_json.get('presence_of_hate', '') hate_degree = response_json.get('hate_degree', '') part_which_contains_hate = response_json.get('part_which_contains_hate', '') foo = response_json.get('fact_or_opinion', '') exp = response_json.get('explanation', '') print(text) print('\n') print("Summary:", summary) print("Sentiment:", sentiment) print("Fact or opinion: ", foo) print("Presence of Hate:", presence_of_hate) print("Explanation:", exp) print("Hate Degree:", hate_degree) print("Part Which Contains Hate:", part_which_contains_hate) return {'title': video_title, 'url': video_url, 'len': video.length, 'lang': result['language'], 'summary': summary, 'sentiment': sentiment, 'foo': foo, 'hate': presence_of_hate, 'explanation': exp, 'degree': hate_degree, 'part': part_which_contains_hate, 'transcript': input_text} except Exception as e: print(f"An error occured: Unable to process")