Non-causal recursive digital filters in multi-band dysperiodicity analysis of synthetic simple vowels

作者：Adnane Mouffak, Mohamed Belbachir · 年份：2012 · DOI：10.1109/setit.2012.6482015 · 被引用次数：1 · 研究领域：Speech and Audio Processing、Advanced Adaptive Filtering Techniques、Speech Recognition and Synthesis

In this paper, we address the application of non-causal two-pass recursive digital filters in special case of multi-band disordered speech analysis, where processing is permissible offline. Most of these methods use linear phase Parks-McClellan optimal FIR filters to decompose speech and its corresponding dysperiodicity into several separate frequency bands. Segmental signal-to-dysperiodicity ratios (SDRSEG) and predictive scores related to each band of analysis are calculated using linear regression and correlated with the original perceived scores. Through this work, we check another alternative of non-causal two-pass recursive digital filtering never used in multi-band acoustic speech processing. Perceptual experiments are performed using corpora of 48 synthetic finite length stimuli of 1 second; including two French vowel categories [a], [u]. These simple vowels cover a range of fundamental frequencies from 100 to 140 Hz, with typical harmonics-to-noise ratio about 20 dB and maximum jitter and shimmer respectively of 1.04% and 3.81%. We show that the uses of this class of non-causal IIR filters via the forward/backward processing can improve method robustness and filters' order comparing to classical Parks-McClellan optimal FIR filtering.