On the effects of filterbank design and energy computation on robust speech recognition

Dimitriadis, D; Maragos, P; Potamianos, A

dc.contributor.author	Dimitriadis, D	en
dc.contributor.author	Maragos, P	en
dc.contributor.author	Potamianos, A	en
dc.date.accessioned	2014-03-01T01:36:33Z
dc.date.available	2014-03-01T01:36:33Z
dc.date.issued	2011	en
dc.identifier.issn	1558-7916	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/21335
dc.subject	Bandpass filters	en
dc.subject	cepstrum analysis	en
dc.subject	error analysis	en
dc.subject	parameter estimation	en
dc.subject	robustness	en
dc.subject	spectral analysis	en
dc.subject	speech processing	en
dc.subject	speech recognition	en
dc.subject	timefrequency analysis	en
dc.subject.classification	Acoustics	en
dc.subject.classification	Engineering, Electrical & Electronic	en
dc.subject.other	Alternative approach	en
dc.subject.other	Alternative energy	en
dc.subject.other	Automatic speech recognition	en
dc.subject.other	Cepstrum	en
dc.subject.other	cepstrum analysis	en
dc.subject.other	Energy estimation	en
dc.subject.other	Energy operators	en
dc.subject.other	Error rate reduction	en
dc.subject.other	Experimental analysis	en
dc.subject.other	Feature sets	en
dc.subject.other	Filter bandwidth	en
dc.subject.other	Filter bank design	en
dc.subject.other	Mel-frequency cepstral coefficients	en
dc.subject.other	Noise conditions	en
dc.subject.other	Noise types	en
dc.subject.other	Noisy recordings	en
dc.subject.other	Noisy speech recognition	en
dc.subject.other	Noisy speech signals	en
dc.subject.other	Relative error rates	en
dc.subject.other	Robust speech recognition	en
dc.subject.other	robustness	en
dc.subject.other	Signal energy	en
dc.subject.other	Spectral analysis	en
dc.subject.other	Speech recognition performance	en
dc.subject.other	Teager-Kaiser operator	en
dc.subject.other	Time frequency analysis	en
dc.subject.other	Bandwidth	en
dc.subject.other	Calculations	en
dc.subject.other	Convolution	en
dc.subject.other	Design	en
dc.subject.other	Error analysis	en
dc.subject.other	Feature extraction	en
dc.subject.other	Filter banks	en
dc.subject.other	Parameter estimation	en
dc.subject.other	Spectrum analysis	en
dc.subject.other	Speech processing	en
dc.subject.other	Speech recognition	en
dc.title	On the effects of filterbank design and energy computation on robust speech recognition	en
heal.type	journalArticle	en
heal.identifier.primary	10.1109/TASL.2010.2092766	en
heal.identifier.secondary	http://dx.doi.org/10.1109/TASL.2010.2092766	en
heal.identifier.secondary	5638124	en
heal.language	English	en
heal.publicationDate	2011	en
heal.abstract	In this paper, we examine how energy computation and filterbank design contribute to the overall front-end robustness, especially when the investigated features are applied to noisy speech signals, in mismatched training-testing conditions. In prior work ("Auditory Teager energy cepstrum coefficients for robust speech recognition," D. Dimitriadis, P. Maragos, and A. Potamianos, in Proc. Eurospeech'05, Sep. 2005), a novel feature set called " Teager energy cepstrum coefficients" (TECCs) has been proposed, employing a dense, smooth filterbank and alternative energy computation schemes. TECCs were shown to be more robust to noise and exhibit improved performance compared to the widely used Mel frequency cepstral coefficients (MFCCs). In this paper, we attempt to interpret these results using a combined theoretical and experimental analysis framework. Specifically, we investigate in detail the connection between the filterbank design, i.e., the filter shape and bandwidth, the energy estimation scheme and the automatic speech recognition (ASR) performance under a variety of additive and/or convolutional noise conditions. For this purpose: 1) the performance of filterbanks using triangular, Gabor, and Gammatone filters with various bandwidths and filter positions are examined under different noisy speech recognition tasks, and 2) the squared amplitude and Teager-Kaiser energy operators are compared as two alternative approaches of computing the signal energy. Our end-goal is to understand how to select the most efficient filterbank and energy computation scheme that are maximally robust under both clean and noisy recording conditions. Theoretical and experimental results show that: 1) the filter bandwidth is one of the most important factors affecting speech recognition performance in noise, while the shape of the filter is of secondary importance, and 2) the Teager-Kaiser operator outperforms (on the average and for most noise types) the squared amplitude energy computation scheme for speech recognition in noisy conditions, especially, for large filter bandwidths. Experimental results show that selecting the appropriate filterbank and energy computation scheme can lead to significant error rate reduction over both MFCC and perceptual linear predicion (PLP) features for a variety of speech recognition tasks. A relative error rate reduction of up to similar to 30% for MFCCs and 39% for PLPs is shown for the Aurora-3 Spanish Task.	en
heal.publisher	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC	en
heal.journalName	IEEE Transactions on Audio, Speech and Language Processing	en
dc.identifier.doi	10.1109/TASL.2010.2092766	en
dc.identifier.isi	ISI:000293702300005	en
dc.identifier.volume	19	en
dc.identifier.issue	6	en
dc.identifier.spage	1504	en
dc.identifier.epage	1516	en