summairpg

README

summairpg - Summarize your role-playing sessions

I was tired of manually writing a summary of the last session of our role-play. So what better to do than use an AI to do it for me? Do you feel the same? Well look no further because this is the open source and free way of doing so! However since GPT-4 is still the best AI on the market you can also use the OpenAI API...

summairpg is just a mere tool to connect two powerful AI frameworks together, namely WhisperX and Ollama (or OpenAI if you prefer).

summairpg follows a three step process:

1. Record audio of the role-playing session. Preferrably provide one audio file per player or game master. If you are using Discord check out Craig to help you with that!
2. Create a full transcription of the role playing session. This project assumes you have setup WhisperX in order to do this.
3. Summarize the transcription using a generative AI of your choice.

Record the voice of all participants

I personally have mostly online role playing sessions using Discord so it's really easy to just add Craig. Afterwards I download the multitrack FLAC bundle which includes one .flac file per user. You can choose another format if you want to save up diskspace but the best results will be with a lossless compression.

Once you have the recordings you need to rename the audio files so they match the name of the ingame character, e.g. 1-myuser-0.flac -> Darell.flac. Put all of these files into one folder with no other audio files.

Creating transcriptions

I choose WhisperX as the speech-to-text tool as it is very fast (even for audio-tracks of several hours) while being very accurate and also providing timestamps on a per-word basis. It fitted all the needs I had.

I personally setup a conda environment for WhisperX that I can just spin up once needed.

Summary via AI

For my own summarization I use the llama3:instruct model as it responds almost instantly with okay results. (I do however have a NVIDIA GeForce RTX 3080 at my disposal...) Still I achieved by far the best results with OpenAIs gpt-4-turbo.

Prerequirements

• WhisperX installed and in PATH (run whisperx --help to check if it works)
• one of these:
  • Ollama installed and serving the HTTP API (ollama serve)
  • OpenAI API Key ready to use

Usage

Provide your audio files in a directory and name them after the role playing characters or GameMaster for the GM. Then you can just run the tool:

./summairpg --audio-dir <dir>

With the default settings this will create a summairpg-config.json file containing your desired configs so if you use the tool again you don't need to provide all parameters again and can just run ./summairpg.

These are all of the available parameters:

$ ./summairpg --help
Usage of ./summairpg:
  -audio-dir string
        The directory that contains all of the audio files that should be transcribed (default "input")
  -audio-display-transcript
        can be set to true to print the entire transcription to console
  -audio-file-types value
        the file extensions that should be considered when looking up audio tracks. They should be a comma-separated list (default flac,wav)
  -audio-language string
        The spoken language in the audio files (default "en")
  -audio-model string
        WhisperX model to use. See https://huggingface.co/models?sort=trending&search=whisper (default "large-v3")
  -audio-transcript-file string
        when set the entire transcription will be skipped and this files content will be used as summarization input
  -config-store
        Store the provided command-line arguments in the summairpg-config.json (default true)
  -ollama-address string
        The host:port of the Ollama HTTP API. (default "127.0.0.1:11434")
  -ollama-enabled
        set to false to disable the Ollama endpoint for summarization (default true)
  -ollama-model string
        Ollama model to use. See https://ollama.com/library (default "llama3:70b")
  -ollama-update-model
        set to false to disable pulling the latest version of the model (default true)
  -open-ai-api-type value
        the type of OpenAI endpoint. Must be one of OPEN_AI, AZURE or AZURE_AD
  -open-ai-api-version string
        the version of the Azure API to use. Not required when openai-api-type is OPEN_AI
  -open-ai-enabled
        set to true to enable the OpenAI endpoint for summarization
  -open-ai-model string
        the OpenAI model to use. See https://platform.openai.com/docs/models/model-endpoint-compatibility (default "gpt-4-turbo")
  -open-ai-org-id string
        will set the OrgID as HTTP header
  -open-ai-url string
        the base url of the OpenAI API endpoint to use (default "https://api.openai.com/v1")
  -openai-api-type value
        the type of OpenAI endpoint. Must be one of OPEN_AI, AZURE or AZURE_AD
  -openai-api-version string
        the version of the Azure API to use. Not required when openai-api-type is OPEN_AI
  -openai-enabled
        set to true to enable the OpenAI endpoint for summarization
  -openai-model string
        the OpenAI model to use. See https://platform.openai.com/docs/models/model-endpoint-compatibility (default "gpt-4-turbo")
  -openai-org-id string
        will set the OrgID as HTTP header
  -openai-url string
        the base url of the OpenAI API endpoint to use (default "https://api.openai.com/v1")