Voice recognition
Speak your incantations into the microphone: automatic install, wake and push modes, confidence, and tips for spelling a Unit ID.
In the Underworld, arts are spoken. With a microphone, Slime Art Online lets you say your incantations out loud: "System Call… Generate Luminous Element… Discharge" — and the orb is born at your fingertip while you speak. Recognition is offline and only knows the words of the incantations, in English (as in the anime, even in the Japanese version).
Installation: nothing to do
On first launch, if voice is not off, the mod downloads by itself what it needs to hear you, once and for all:
| Piece | Size | Source |
|---|---|---|
| Vosk library | ≈ 26 MB | Maven Central, checked against a fingerprint |
| JNA library | ≈ 2 MB | Maven Central, checked against a fingerprint |
| Small English model | ≈ 41 MB | official Vosk site (alphacephei.com) |
The chat keeps you posted: "Slime Art Online is fetching offline speech recognition for your incantations (about 67 MB, once).", the progress by quarters, then "Your incantations can be heard now. Say "System Call".".
- The download runs in the background: it never blocks the game from starting.
- Everything is installed in your instance's
slimeartonline-voice/folder (lib/andmodel/subfolders). - Nothing leaves your computer: recognition runs locally, your voice is sent nowhere.
- Nothing is downloaded if
mode = "off"orautoDownload = false.
The three modes
The mode setting in slimeartonline-client.toml (see Configuration):
| Mode | The microphone listens… | For whom |
|---|---|---|
wake (default) |
all the time, but only for incantations; saying "System Call" opens one | hands-free play |
push |
while you hold the Incantation: push to talk key (and 0.7 s after) | streaming, playing on voice chat with others |
off |
never | no microphone; nothing is downloaded |
The push key is unbound by default: assign it in the controls (Controls).
How it hears you
- A closed vocabulary: the microphone only looks for incantation words. Everything else — your conversation, music — is filed as noise and ignored. In
wakemode, nothing fires until you have said "System Call" (or a weapon art, weapon in hand). - As the sentence goes:
Generate … Elementcommands fire while you are still speaking, as soon as they are heard in full and stable for a quarter of a second. AGenerateat the end of a sentence waits a moment, in case "Ten Times" follows. - Decisive words wait for the end of the sentence:
System Call, the closing words (Discharge,Burst Element,Adhere) andTraceonly fire once the sentence is over and checked word by word against the confidence threshold. - What it hears is displayed in grey, between quotes, on the left of the screen — in the incantation console during an incantation.
Tuning confidence
confidence (0.55 by default, from 0 to 1) is the threshold below which a heard word is discarded.
| Symptom | What to do |
|---|---|
| Incantations trigger by themselves | raise the confidence (0.65 – 0.75) |
| Your commands are often ignored | lower it slightly, speak more clearly, bring the microphone closer |
Spelling a Unit ID
For a Trace, spell the letters in the radio alphabet, then the digits one by one in English:
Trace, Unit ID, Delta Charlie Hotel, Two Zero Four Six One
| Letters | Digits |
|---|---|
| Alpha, Bravo, Charlie, Delta, Echo, Foxtrot, Golf, Hotel, India, Juliet, Kilo, Lima, Mike, November, Oscar, Papa, Quebec, Romeo, Sierra, Tango, Uniform, Victor, Whiskey, X Ray, Yankee, Zulu | Zero, One, Two … Nine |
- You can catch your breath: an incomplete Trace opens a 15-second spelling window (a small click signals it, and the console shows
TRACE UNIT IDwith a cursor). Following sentences made only of radio letters and digits complete the ID; each piece restarts the 15 seconds. - The start is often enough: as soon as what you have spelled (at least three letters) matches a single creature loaded around you, the ID completes itself. "Trace, Delta Charlie Hotel" is often enough.
- In chat, the ID is written in one piece:
trace unit id DCH20461.