The habits that make Taperat noticeably smarter, and a checklist for your first 15 minutes.
Before importing your entire collection of recordings, try a game or two. Transcribe them, or run the Perception Engine on them, to get to know the layout of the app. If you have used a video editor like DaVinci Resolve or Premiere Pro, you will feel right at home.
After every Perception Engine run you get a summary, where you can correct speaker names and scene findings, and look through the transcript and the diarisation (who spoke when). Don’t skip it, and where you can, fix the outliers.
When it asks “Is Speaker 2 Elara?”, a quick Keep or Not this teaches it more than any setting. It will save you hours of work later.
Voice fingerprints get better the more you teach them. Start with scenes where fewer than four characters talk, and name the speakers there. Every correction helps it recognise those voices later.
Once a voice has a fingerprint, Taperat picks it out of bigger groups too, and everything else improves with it.
Play with subtitles on if the game offers it, ideally with speaker names shown. Subtitles help fix misheard words and make names on screen searchable.
Don’t want to play the whole game like that? Subtitles on a handful of clips already help.
The first time you run the Perception Engine on a game, Taperat asks what kind of game it is, so it knows what to look for. It suggests an answer from your recordings, and you can change it any time on the Games page in your library.
Speech on your mic track suggests Multiplayer; subtitles on screen suggest Story.
If the game lets you, turn music and effects down a little and keep voices up. Clear speech means cleaner transcripts and stronger fingerprints.
Let an AI agent go through your recordings for you. Through MCP Bridge it searches transcripts, on-screen text and scenes, then creates slices, markers, tags, collections and sessions, or starts transcription and Perception Engine runs: whatever you ask of it.
Using an external agent locally? Great! You can connect any agent you like.
“In this session from game X, make slices of every battle and every story beat with character Y, and drop a marker at the start of each chapter. Separately, find the scene where character Z betrays character Y. Put them in the Story and Combat collections and tag the characters.”
Connecting an external agent or AI assistant that runs in the cloud shares your files with that service. For footage under NDA, keep MCP Bridge off or use an assistant that runs on your PC.
Clearly named and tagged slices and recordings let your agent find exactly what it needs and place it straight onto your editor’s timeline, in DaVinci Resolve Studio or Premiere Pro: on a new timeline or at the playhead, with the exact in and out points.
So give slices clear titles and put them in collections as you go. Good names help your own searches just as much.
“Put every slice from the Combat collection with character Y on a new timeline, in the order they happened.”