ZundamonGPTonYouTube

English | 日本語

Intelligent Zundamon replies YouTube chat with GPT brain.

First of all

• This Application is Japanese only since it's depend on Japanese voice engine "VOICEVOX", but You can cutomize by modifying MIT liccenced source codes.

This application works on

• Windows OS (tested on Windows 10)

• .Net Framework v.4 (tested on v4.7.2)

• the machine on which installed VOICEVOX (tested on v.0.14.6)

  Core Module is implemented by Python, so it can adapt to other OS or voice generators.
  
  

This application can

• automatically pick up messages from YouTube chat and make Zundamon speak the GPT answer of those messages out.
  Thogh non Japanese messages are given, Zundamon answers in Japanese.
• display all comments of YouTube chat, picked up comments, answers of picked up comments.
• display Zundamon portrait with transparent background.
• You can use not only Zundamon voice and image but also other ones.
  

Usage

• Install VOICEVOX
• Get OpenAI api-key. Please refer here(English) or here(Japanese)
• if you want to launch from .exe file.
  • click here to download newest version.
  • Unzip Downloaded "ZundamonGPTonYouTube.zip" file.
  • Open "ZundamonGPTonYouTube" and double click ZundamonGPTonYouTube.exe.
• if you want to launch from source codes.

  • Install ffmpeg.
    For Linux: Execute following command.

    $ sudo apt-get install ffmpeg
    

    For Windows: Access here, download '*-win64-gpl.zip', extract the zip and move three exe files (ffmpeg.exe, ffprobe.exe, ffplay.exe) to the folder where you'll execute the sample or added path.
    
    For Mac: Access here, copy installation command to your terminal and push Enter key, and execute following command.

    brew install ffmpeg
    

  • Clone repository.
    

    git clone https://github.com/GeneralYadoc/ZundamonGPTonYouTube.git
    

  • Move to ZundamonGPTonYouTube directory

    mv ZundamonGPTonYouTube.
    

  • Install the application.

    pip install .
    

  • Start the application.

    python3 ZundamonGPTonYouTube.py
    

• Check Video ID of target YouTube stream.
  
• Fill in the Video ID brank of start form. (use Ctrl+V to paste)
• Fill in the API Key (of OpenAI) brank of start form. (use Ctrl+V to paste)
• Click "すたーと" button which means "start".
  

Notice

• OpenAI api key and Video ID is recorded in "variable_cache.yaml" and You can skip either or both from the 2nd time.
• Please be aware of treating "variable_cache.yaml" in order to avoid leaking OpenAI api key.
  
  

GUI is consisted of

Main window

• You can change visibility of chat window by pressing "ちゃっと" button, asking window by pressing "しつもん" button, answering window by pressing "こたえ" button, portrait window by pressing "立ち絵" button.
• You can change voice volume by using slide bar which is at the bottom of the window.
• Also you can change voice volume by putting value on text box at the just right of the slide bar and press enter key.
• You can exit the application by closing this window. window "x" button of TopRight.
  

Portrait window

• You can display a portrait of avatar which you like by specifying the path in setting file.
• You can switch opaque or transparent background by double clicking the avatar.
• You can resize the avatar in opaque background mode, please erase background after adjusting avatar size if you want.
• Minimizinq window also is available in opaque backgroune mode.
  
• The Application keeps running even if this window is closed, so you can close this window if unnecessary.
  

YouTube chat monitor window

• Almost all messages are shown in this window.
• Messages contain only emoticons are ignored.
• Some messages which exist in polling gap may lost.
• You can switch visibility of window frame by double clicking message area of the window.
• Please turn on the frame when resizing the window.
• The Application keeps running even if this window is closed, so you can close this window if unnecessary.
  

Window for asking

• All picked up messages which will be answered by ChatAI are shown in this window.
• You can switch visibility of window frame by double clicking message area of the window.
• Please turn on the frame when resizing the window.
• The Application keeps running even if this window is closed, so you can close this window if unnecessary.
  

Window for answering

• ChatAI answers for picked up messages are shown in this window.
• You can switch visibility of window frame by double clicking message area of the window.
• Please turn on the frame when resizing the window.
• The Application keeps running even if this window is closed, so you can close this window if unnecessary.
  
  

Notice

• The following window is VOICEVOX which is external application.
  It's necessary for generating Zundamon voices, so please don't close the window. (please minimize if you want to hide it.)
  
  
  
  

Settings

You can customize the application with "setting.yaml" which is exist in the same layer of the application exe file.

# VoiceVoxの設定
voicevox_path: ''

# チャット欄ウィンドウの設定
display_user_name_on_chat_window: true
chat_window_title: 'ちゃっとらん'
chat_window_padx : 9
chat_window_pady : 9
chat_window_color: '#ffffff'
chat_font_color: '#000000'
chat_font_size: 10
chat_font_type: 'Courier'
chat_rendering_method: 'normal'

# 質問ウィンドウの設定
display_user_name_on_ask_window: false
ask_window_title: 'ぐみんのしつもん'
ask_window_padx : 9
ask_window_pady : 9
ask_window_color: '#354c87'
ask_font_color: '#ffe4fb'
ask_font_size: 12
ask_font_type: 'Courier'
ask_rendering_method: 'refresh'

# 回答ウィンドウの設定
answer_window_title: 'てんさいずんだもんのこたえ'
answer_window_padx : 9
answer_window_pady : 9
answer_window_color: '#ffe4e0'
answer_font_color: '#004cF7'
answer_font_size: 13
answer_font_type: 'Helvetica'
answer_rendering_method: 'incremental'

# 立ち絵ウインドウの設定
image_window_title: '立ち絵'
image_window_refresh_rate: 30
image_window_transparent_color: '#00ff00'
image_window_font_color: '#0000ff'
image_window_font_size: 11
image_window_font_type: 'Helvetica'
image_window_label: 'ダブルクリックで\n背景透過/非透過を\n切り替えられます'

# AIの設定
model: 'gpt-3.5-turbo'
max_tokens_per_request: 1024
ask_interval_sec: 20.0

# 回答キャラクターの設定
speaker_type: 1
volume: 100
system_role: 'あなたはユーザーとの会話を楽しく盛り上げるために存在する、日本語話者の愉快なアシスタントです。'

• "voicevox_path" can remain blank if VOICEVOX has been installed to default path.
• You can change AI model by changing "model" value.
• You can change voice actor by changing "speaker_type" value.
• You can change Avatar image by changing "image_file" path.

Current size and position, frame visibility, bagckground transparency of these windows is memorized and inherited to them in next time of executing.
They are recorded in "variable_cache.yaml". you can change window size and position also by editing the file when the application is not running.

answer_frame_visible: false
answer_window_height: 450
answer_window_visible: true
answer_window_width: 500
answer_window_x: 659
answer_window_y: 521
api_key: XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
ask_frame_visible: false
ask_window_height: 250
ask_window_visible: true
ask_window_width: 500
ask_window_x: 663
ask_window_y: 224
chat_frame_visible: false
chat_window_height: 754
chat_window_visible: true
chat_window_width: 350
chat_window_x: 246
chat_window_y: 225
image_bg_visible: false
image_window_height: 816
image_window_visible: true
image_window_width: 522
image_window_x: 1234
image_window_y: 175
video_id: XXXXXXXXXXX

Licence

• The lisence type of this application is MIT, so you can customize freely.
• the lisence type of ffmeg executable files included in release package is LGPL.
  
  

Links

• Pixiv page of 坂本アヒル   I obtained static Zundamon portrait which is the material of the gif animation from here.
• ChatAIStreamer   Python library for getting ChatGPT voiced answer of YouTube chat stream.