Dungeon Scribe

As a DM who runs very long sessions, usually 4-8 hours, taking notes has always been a challenge for me. Eventually I started recording them and realizing I didn't really have the time to hunt through audio files, I had the idea to transcribe them. 

So, Dungeon Scribe is my solution. 

What's it do? In short, you feed it audio and it transcribes it, summarizes it, and then spits it out to Discord or Obsidian.

After the transcript is created, you can optionally use an online AI service (like Google Gemini or OpenAI's ChatGPT) to generate a summary of the session. In my testing, the API cost there is pennies - and I haven't yet found a clear winner between the two.

The main goal here is to provide a  record of your campaign without needing to take extensive manual notes during play. It's also quite nice to have the summary in Discord.

Features

• Audio-to-Text Conversion: Converts common audio files (.mp3, .wav, .m4a) into a text transcript.
• Hardware Acceleration: Can use your NVIDIA or AMD graphics card to speed up transcription. A CPU-only option is also available.
• AI Summaries: After creating a transcript, you can connect to Google Gemini or OpenAI (requires your own API key) to generate a summary of the session.
• Custom Dictionaries per Campaign: Create separate profiles for each of your campaigns. You can add lists of character names, locations, and unique game terms to improve the accuracy of the transcript for that specific campaign.
• Built-in Audio Recorder: Record directly from a microphone or capture your computer's audio output (useful for online sessions on Discord).
• Session Management (beta): Automatically saves all transcripts and summaries, organized by campaign and session number, for easy access later.
• Export Options: Save your transcripts or summaries as plain text files, send them directly to a Discord channel, or create a new note in an Obsidian vault.
• Large Audio File Handling: Automatically splits very long audio files into smaller pieces during processing to ensure stability and prevent crashes.
• Speaker Identification (Experimental): An optional feature that attempts to label different speakers in the transcript (e.g., "speaker 0," "speaker 1"). This feature is experimental and may be less accurate. I haven't had a lot of luck with this yet - sometimes it works awesome but most of the time it doesn't.