This small-scale continuous voice recognition technology is optimized to embedded environments so as to enable voice recognition even in a state without a network connection. Working successfully even with low-end/ low-power CPU, Always-ON ASR controls devices anytime, anywhere regardless of distance through only using their voice.
Key algorithm
The continuous voice recognition algorithm that is powerful for noise removal and is based on call command for long-distance always-on voice recognition systems that enable users to control various smart electronic appliances anytime, anywhere through only using their voice.
- Powerful call command-based continuous voice recognition technology
- Optimized for embedded environments with low price/ low power consuming/ small capacity memory
- No touch! Button-less remote voice control/ always-on type
- Unlimited multilingual support
- 80MHzCPU/ 2MB memory
- Supporting Windows/ CE/ Embedded Linux/ RTOS/ Android/ iOD
- Out of vocabulary rejection technology
- Optimal ada-Boost system
- GMN/ SVM-based classified design
- LM model design based on n-gram
- Real-time embedded speaker adaptation technology
- Technology to optimize user’s recognition performance
- Although this is basically a speaker-independent voice recognition system, it improves recognition rate by
taking into consideration the characteristics of user’s voice and the surrounding environment. - The technology for learning the speaker’s voice characteristics is designed for optimization so as to be applied
to noise removal in low-end CPUs. - Speaker adaptation technology of probability statistics based on map algorithm
