Playable demos & bots
SpellSpeak Tone reads each line of game dialogue and tags how the speaker feels, how the line lands on everyone present, and whether it asks, answers or dodges.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
It is a 28M-parameter MiniLM classifier rather than a general decision model, built so several voiced characters can react to a line in about 10 ms on a CPU. Jon Taylor's Pipecat demo uses it to track how characters feel about the player. Its maker warns against using it for moderation or for judging real people.
In a game or story with several AI characters, Tone can tell each character whether the player just thanked, teased or threatened them, so faces, moods and who speaks next can change straight away. It reads English only and misses some sarcasm and hostility, so treat its tags as cues for characters, not as verdicts about real people.