Start from a published base model, tune the vocabulary to your task, then bake a fresh .rivlet binary + cross-platform bundle. ← back to /inference
GSCv2 — the canonical 35 short spoken words. The default for voice-controlled devices, kiosks, and accessibility tooling.
Tiny custom wake-word base. Add 1-2 phrases, deploy on a Raspberry Pi as a 'hey atlas' / 'computer' listener.
Doorbell, knock, dog bark, glass break, alarm, baby cry, vacuum, water running. For smart-home situational awareness.
Motor hum, bearing fault, pump cavitation, leak hiss, conveyor belt slap. A predictive-maintenance signal source.
Zero through nine plus 'cancel' and 'enter'. For voice-controlled keypads, phone trees, smart-meter readouts.
Start with no preset vocabulary. Upload your own audio, define your own classes — full custom training from the ground up.