Launching today
Table Canon
Automated notes, searchable history, prep ideas for GMs
54 followers
Automated notes, searchable history, prep ideas for GMs
54 followers
Focus on the session, not note taking. Simplify your prep time with a searchable memory-bank of what happened at your table. Table Canon turns uploaded game-night audio into a recap, outline, entity dossiers, searchable index, and prep ideas for tabletop RPG game masters. Memory evolves each session. No sign up required to get started, 6 hours of audio upload free! Message me that you saw the launch on Product Hunt and I’ll credit your account with an extra 10 hours.
Free Options
Launch Team / Built With
Customer.ioAutomate Messaging Everywhere — Startups Get 12 Months Free
Promoted
Hi Derek, can you share more about what the Audio Worker works? I'm assuming it is using AI to do the transcription. Also, can the GM upload more details in advance of uploading the transcription to assist the AI?
Good luck with today's launch!
@sara_ford_goog
Thanks for the question!
The Audio Worker is what picks up a session after the recording is uploaded. It splits the audio into overlapping chunks and runs each one through a custom transcription and diarization pass (Transcription is Whisper large-v3, through WhisperX, and the voice separation is pyannote 3.1. If someone at the table has recorded a short voice sample, we match speakers with a SpeechBrain ECAPA voiceprint.), so yes, transcription is AI in that sense. That pass writes down what was said, with timestamps, and a second model separates the voices so lines can be tied to different speakers. After the transcript exists, another pass reads it and builds the campaign record from it: who was talking, the people and places that came up, and the session recap.
The GM can set a lot of that up before the audio goes in. On the campaign desk there is a glossary for names, places, and homebrew terms. Those get passed to the transcriber so it spells your NPCs and locations the way the table actually says them, instead of guessing. Adding the party helps too, and a short voice sample makes speaker matching much better, though it is not required to upload. Characters and places already in the campaign are used again when the recap is written, so the model can match what it heard to what you wrote down ahead of time.
Happy to go deeper on any of that if you're interested!