Conferences and supporting programme
Ultra Low Power Key Phrase Detection at the Edge
With increased interaction with smartphones, tablets & personal assistance systems, interacting with technology using voice is becoming one of the preferred methods of human machine interaction. Today users can ask Alexa to order things for them, automate their homes controlling lights, thermostat, door locks, & garage doors or even help water their lawn. This type of interaction has become the norm. Solution are available to increase the interaction with technology using highly accurate, small in size, low cost and power efficient flexible Neural Network (NN) based key phrase detection solutions. Unlike the cloud connected NN key phrase detection implementations with the need for internet connectivity and concerns regarding security and privacy, edge based solution such as the one discussed on this paper does all computing at the edge and does not record or transmit data to the cloud. A Binarized (BNN) or 16 bit Convolutional Neural Network (CNN) models are available and run on UltraPlus low power FPGAs. Key phrase detection can be done even in noisy environment when the NN is trained with dataset that include noisy background such as music or chatting noise. The NN in this case is trained using a public dataset to detect the word “seven”. Key phrase detection can be utilized in wide range of applications without the need for a personal assistance device. Possible applications are smart light switches, smart TVs or AVRs managing devices with commands such as volume up/down.
--- Date: 26.02.2019 Time: 10:30 AM - 11:00 AM Location: Conference Counter NCC Ost