Archivar Horst AI
A Discord bot with speech input and output. Follows six voice channels at once, attributes what is said to speakers and answers as speech. Plus its own data on the guild and a scheduled reconciliation with external leaderboards.
- Status
- Live
- Access
- Usable without signing in Profile page is open, the bot itself lives in the guild's Discord
- Stack
- Node · Discord · Spracherkennung · Sprachausgabe · MySQL
AI at work Data storage External service
Short-lived notes expire after two days, facts stay. Without that split the store would be empty or unusable within weeks.
Following six voice channels at once shaped the architecture. Transcription runs continuously rather than on demand and has to attribute what is said to individual speakers, otherwise the store fills up with text nobody said.
On the output side the real problem was how numbers are pronounced. Years came out slurred into a single word, so I insert pauses between the parts before the text goes to synthesis.
The memory has two half-lives, and it had to. Keep everything and the store is unusable within weeks; discard everything and it recognises nobody. Short-lived notes therefore expire after two days, facts stay.
The external sources - leaderboards, ratings, combat logs - are polled on a schedule. The work is not in fetching them but in reconciling them with the local data.
The profile page is public, the bot runs in the guild’s Discord.