A 510-nW Wake-Up Keyword-Spotting Chip Using Serial-FFT-Based MFCC and Binarized Depthwise Separable CNN in 28-nm CMOS.
Weiwei ShanMinhao YangTao WangYicheng LuHao CaiLixuan ZhuJiaming XuChengjun WuLongxing ShiJun YangPublished in: IEEE J. Solid State Circuits (2021)
Keyphrases
- keyword spotting
- speech recognition
- cmos technology
- nm technology
- metal oxide semiconductor
- silicon on insulator
- low power
- power consumption
- analog vlsi
- speech processing
- low cost
- document images
- printed documents
- hidden markov models
- speech signal
- high speed
- power dissipation
- low voltage
- speaker identification
- language model
- cmos image sensor
- integrated circuit
- image sensor
- floating point
- automatic speech recognition
- single chip
- ibm power processor
- random access memory
- feature extraction
- circuit design
- handwritten documents
- gaussian mixture model
- pattern recognition
- input image
- noisy environments
- dynamic range
- chip design
- mixture model
- signal processing
- neural network