The Keyword Spotting(KWS) system provides a balance of convenient wake-up and energy-saving for the voice assistant, but system security and user privacy are still difficult to ensure. This technology integrates speaker recognition into KWS recognition system with ultra-compact deep neural network classifier implemented on microcontroller(MCU). Our systems are superior to existing systems in both computational efficiency and memory requirements.

This technology integrates the speaker recognition into keyword spotting(KWS) recognition, and implements a secure KWS system on low cost microcontroller. In addition to the higher recognition performance and better system security advantages of our system, this technology has lower computational complexity and memory requirement than existing KWS systems.

This technology is applicable to all human-computer interaction systems that require voice-triggered activation, such as popular voice assistant products such as Google Assistant, Apple＇s Siri, Microsoft＇s Cortana, and Amazon＇s Alex. In addition to improving the security of the system, the low computational complexity and memory requirement are more conducive to achieving low power, low cost microcontroller-based product design.