EchoPenalty (subtitles & banned words)

EchoPenalty transcribes your microphone with a small speech model (whisper.cpp) and either shows subtitles in the overlay or plays "EchoPenalty": you must not say certain words — if you do, you lose HP. It shows an HP bar, hearts and a time-based shield that absorbs one mistake; at 0 HP it shows "Game Over". Speech recognition runs entirely on your machine, no audio is sent to the internet. A dedicated Quick Control section allows start/stop/reset with a live HP bar. For automations, alongside "forbidden word said", "HP changed" and "game over" there is also a "shield activated" trigger; a ready-made "EchoPenalty Game Over → Discord" template is available in the automation gallery. Since 1.0.15 the microphone and speech recognition only run while subtitles are on or a game is running. The automation action "Toggle subtitles" really switches subtitles on and off. Start/stop is now also available as a Stream Deck and phone remote key. Since 1.0.24: the penalty spin is queued if the random wheel is already spinning; if it is off, the tab shows “Wheel not found”. An empty wheel name uses the last used wheel. Since 1.0.25: Optional chat messages (off by default, custom texts with placeholders): on a banned word (word + remaining HP) and on game over. New in 1.0.28: After game over the microphone is closed, and hit messages in chat appear at most every 5 seconds.
Trigger Runs as soon as the module is enabled. Start the game with the "Start game" button in the module tab (or via automation).
Requirements
- Premium – or the free 14-day Premium trial (see “Premium trial”).
- A microphone that records you (not the game/guest audio).
- A one-time model download in the tab (tiny ≈ 75 MB, base ≈ 140 MB, small ≈ 470 MB) — not bundled with the installer.
- Some CPU headroom: transcription runs in its own process; load can be capped via model size and thread count.
Step by step
- Open the "EchoPenalty" module tab (Games category) and enable the module.
- Under "Download model" pick a size — recommended: base. tiny is faster but less accurate; small handles mumbling better but costs more CPU.
- Select your microphone from the list. If the list is empty, enable the module briefly so Windows grants the permission, then click "Refresh".
- Choose a mode: "Subtitles" shows the spoken text as a caption box. "Game (EchoPenalty)" enables the HP game.
- For the game: add banned words as a list. Matching is either "word boundary" (exact only) or "substring"; umlauts are normalized.
- Optionally set max HP, HP loss per word, shield charge time and — if desired — "spin wheel on banned word" (spins your existing prize wheel).
- Position via the Positions tab, look via colour presets (free) or your own AI design (premium).
- Start the game with "Start game"; "Reset" restores HP.
If it doesn't work
- No subtitles: check the status in the tab ("Running") and that the correct microphone is selected. On "Error", download the model again.
- Words arrive with a delay: this is normal (1–3 seconds). A smaller chunk length is faster but less accurate.
- Banned words get missed: pick a larger model (small) or switch individual words to "substring".
- High CPU load: reduce the thread count, pick a smaller model or raise the silence threshold slightly.
- Guest/game audio gets transcribed: in Windows sound settings make sure the selected input device only captures your voice.
Last updated: 15 November 2026