Trigger Word

is the trigger word detection engine for CogniTech. It is designed to operate with or without an Internet connection. It could work in fully standalone mode (for devices not connected to the cloud), or it could trigger a powerful speech recognizer located on the cloud (such as Alexa or OK Google) or on a desktop environment. Cognit operates in real-time mode, has very small memory footprint  and low latency thus perfectly suited for embedded applications, which usually require always-on status. The system could be used in a wide range of applications whether involving  stand-alone devices/machines or systems that are tethered to Local Area Networks (LAN) or to the Internet such as Internet of Things.

Cognit provides developers/customers with the ability to define their own trigger word. They don’t need to train it with sample utterances as is the case with a number of existing commercial offerings. CogniTech has developed a technology (patent pending), which searches for and adequately concatenates selected bits of sounds extracted from thousands of hours of collected voice recordings in various accents and multitude of environments (quiet, noisy, and in-between). CogniTech provides currently Trigger world in English, and is working on expanding the offering to a variety of other languages. The model has been developed and tested by CogniTech using powerful Deep Learning technology and AI based tools.  

Cognit provides developers and third party users with models for the trigger word they select. It supports cross platform implementation and has been tested on variety of desktop/server operating systems including Linux, Windows 10, Remix, IOS, Android;  and on large number of embedded platforms, including  Jetson CPU/GPU; Pine64 Debian CPU; Raspberry Pi CPU, to name a few.