pplying hybrid “CD-CNN-HMM” model for keywords spotting in continuous speech

摘要

Recently, there has been an increased interest in Keywords Spotting (KWS) systems. They have been widely used in various applications, like, spoken data retrieval, speech data mining, spoken term detection, telephone routing, etc. These applications require low computational cost and high precision. To meet these requirements, we propose in this work a systematic approach of keywords spotting (KWS) in continuous speech. This system performs on two-stage, in first one the continuous speech is decoded into phonetic flow using a Context Dependent-Convolutional Neural Network-Hidden Markov Model (CD-CNN-HMM) built with the open source speech recognition toolkit Kaldi, and in the second stage the keywords will be identified and detected from this phones sequence using the Classification and Regression Tree (CART) implemented with the software MATLAB. The work will be conducted on TIMIT data set.