Wavelet – Neural Data Mining Approach for Spoken Keyword Spotting
- K. A. Senthil Devi
- DR. B. Srinivasan
Spoken keyword spotting, Speech data mining, Wavelet Packet Decomposition, Discrete Wavelet Transformation, BPN neural network, word detection
Spoken keyword spotting is a technologically relevant problem in speech data mining. It is essential to identify the occurrences of specified keywords expertly from lots of hours of speech contents such as meetings, lectures, etc. In this paper, Wavelet Packet Decomposition (WPD) and Neural Network (NN) based data mining model (WPDNNM) is explored for keyword spotting. Speech data is first decomposed with Haar, Daubechies2 and Simlet4 wavelets packets. Then, some significant features are extracted from the decomposed speech data. Back Propagation Neural Network (BPNN) is trained with three predefined spoken keywords based on known features and finally, input speech features are compared with keyword features in the trained BPNN for spotting the occurrences of the specified keyword. The method of this paper is tested with 5 minutes lecture data. This method is compared with Discrete Wavelet Transformation (DWT) feature extraction based keyword spotting. Experimental results show that the wavelet - neural method with WPD of Daubechies2 wavelet is more accurate than with Haar and Simlet4 wavelets.
K. A. Senthil Devi, DR. B. Srinivasan. "Wavelet – Neural Data Mining Approach for Spoken Keyword Spotting".INTERNATIONAL JOURNAL OF ENGINEERING DEVELOPMENT AND RESEARCH ISSN:2321-9939, Vol.5, Issue 1, pp.569-576, URL :https://rjwave.org/ijedr/papers/IJEDR1701086.pdf
Volume 5 Issue 1
Pages. 569-576
