installation

1. install python >= 3.9 <3.12 (TTS required) and docker https://www.python.org/downloads/release/python-31010/ https://www.docker.com/get-started/

under windows paltform run the install.bat it will run the following commands pip3 install -r requirements.txt winget install ffmpeg docker compose up docker exec ollama ollama pull llama3:8b docker exec ollama ollama pull ycchen/breeze-7b-instruct-v1_0

2. use following commands to install packages in requirements.txt pip3 install -r requirements.txt

3. install FFmpeg linux: sudo apt install ffmpeg

windows: winget install ffmpeg

macos: brew install ffmpeg

4. install ollama(LLM) and ttrss automatically through docker compose docker compose up docker exec ollama ollama pull llama3:8b docker exec ollama ollama pull ycchen/breeze-7b-instruct-v1_0

optional (suubscribe rss)

1. get rss subscribe link

can easily get the subscribe link by rsshub radar chrome extension

https://chromewebstore.google.com/detail/rsshub-radar/kefjpfngnndepjbopdmoebkipbgkggaa

the subscribe link format:(e.g. nature vedio: https://www.youtube.com/@NatureVideoChannel) https://www.youtube.com/feeds/videos.xml?channel_id=UC9bYDXoFxWC2DQatWI366UA

2. subcribe some youtube channel through ttrss

the GUI interface of ttrss: http://localhost:181/

user：admin password：password

or

user：admin password：ttrss

(can set the user and password through compose file or change it in GUI after login)

You can find the subscribe button by clicking on the top-right menu button.

documentations for ttrss: https://ttrss.henry.wang/zh/#%E9%80%9A%E8%BF%87-docker-%E9%83%A8%E7%BD%B2

3. get the rss feed link from ttrss through orange rss icon. the format is like : http://localhost:181/public.php?op=rss&id=-3&is_cat=0&q=&key=2wje2q662703f6266c7

Project Analysis and Future To-Dos for Video Summarization Script

Function Definition

Function Description This script automates the entire process of extracting, cleaning, and summarizing textual information from YouTube videos or ttrss RSS feeds. The final output includes summarized text in document format and optionally synthesized speech audio files.

Main Functions

1. Argument Parsing:

   • Uses argparse to parse command-line arguments to control the script's behavior.
   • Supported parameters include YouTube or ttrss link, language, model name, output directory, etc.

2. Download and Subtitle Generation:

   • Uses yt-dlp to download videos or extract audio.
   • Generates subtitles using Whisper.
   • Cleans subtitle files by removing timestamps and other non-content elements.

3. Content Summarization:

   • Splits subtitle text into chunks and sends them to a large language model (LLM) for summarization.
   • Combines and refines the summarized text.

4. Document and Audio Generation:

   • Creates .docx files containing summarized text.
   • Optionally generates TTS audio files from the summarized text.

Future To-Dos

1. Code Improvement and Optimization:

   • Modularization: Separate main functions into independent modules or classes to improve code maintainability and reusability.
   • Error Handling: Enhance error handling mechanisms to better manage failures in downloading, subtitle generation, etc.

2. Feature Expansion:

   • Multilingual Support: Add support for more languages, especially in the TTS generation and LLM summarization parts.
   • Multi-Platform Support: Ensure the script runs seamlessly on different operating systems (e.g., Windows, MacOS, Linux).

3. Performance Optimization:

   • Parallel Processing: Introduce multi-threading or multi-processing to speed up video downloading, subtitle generation, and summarization.
   • Caching Mechanism: Add caching to avoid reprocessing the same video or audio files.

4. Testing and Documentation:

   • Unit Testing: Add unit tests for each major function to ensure code reliability and maintainability.
   • Usage Documentation: Write detailed usage instructions and API documentation to help users and developers understand and use the script.

Detailed Function List

1. Argument Parsing:

   • Parse command-line arguments.
   • Supported arguments: link, prompt, language, Whisper model size, model name, timestamp content, output directory, picture embed, TTS creation.

2. Download and Subtitle Generation:

   • Handle YouTube videos and ttrss links.
   • Download videos and extract audio.
   • Generate subtitles using Whisper.
   • Clean subtitle files.

3. Content Summarization:

   • Split subtitle text into chunks.
   • Summarize content using LLM.
   • Combine and refine summarized text.

4. Document and Audio Generation:

   • Create .docx files.
   • Optionally generate TTS audio files.

usage

the script support a cli-interface:

('--link', type=str, help='The link to the youtube video,or a ttrss link')

('--prompt', type=str, default="summary the following content", help='template to llm, in en or zh')

('--language', type=str, default= "en", help='language of output')

('--whisper_model_size', type=str, default='medium', help='model to use for whisper')

('--model_name', type=str, default="auto", help='llm model,in ollama format:https://ollama.com/library')

('--timestamp_content', type=str, default="False", help='use content contains timestamp in LLM inference')

('--output_dir', type=str,default= script_dir, help='output directory')

('--pic_embed', type=str, default='True', help="True or False, decide whether to embid scrrenshot in docx file")

('--TTS_create', type=str, default='True', help="True or False, decide whether to perform TTS")

process single youtube video

python3 /mnt/d/youtubescript/yt_transcript.py --link "https://www.youtube.com/watch?v=0RtcsA5MQPs" --temp "give me three key points in the context" --language en

process youtube video in playlist

python3 /mnt/d/youtubescript/yt_transcript.py --link "https://www.youtube.com/playlist?list=PLOAQYZPRn2V5RA1URMVkHFQYgLrkuXST9" --timestamp_content True --temp "give me three key points in the context and the timestamp" --language en --output_dir /mnt/d/youtubescript/Vivian_pop_sharing

process each video record in ttrss link

python3 /mnt/d/youtubescript/yt_transcript.py --link "http://localhost:181/public.php?op=rss&id=-3&is_cat=0&q=&key=2wje2q662703f6266c7" --temp "從內容中萃取出一個商業機會" --language zh