WSEAS Transactions on Signal Processing
Print ISSN: 1790-5052, E-ISSN: 2224-3488
Volume 11, 2015
Low-Power OZGF Bank and MR Hamming Windowing for Embedded Speech Recognition
Authors: , ,
Abstract: We present novel implementations of a One-Zero Gammatone Filter and a multiresolution Hamming Window with constant time complexity for low power digital implementation in embedded speech recognition systems. We compare our model with state-of-the-art basilar membrane models in terms of computational complexity and in terms of phone classification accuracy on the TIMIT dataset and show quantitative advantages in both, enabling better speech recognition for a broader class of power and resource constrained digital embedded systems.