On data-derived temporal processing in speech feature extraction

Abstract

Temporal processing and filtering in speech feature extraction are commonly used to aid in performance and robustness in automatic speech recognition. Among the techniques successfully employed are RASTA filtering, delta calculation, and cepstral mean subtraction. The work here explores the use of temporal filter design using LDA to further enhance performance using a few preprocessing configurations. In addition to RASTA filtering, we apply the filters to modulation-spectral features and cepstra while making sure that the assumptions of LDA are observed. We additionally test the use of filters that have been trained in different reverberation conditions, noting from previous work that the presence of reverberation alters the preferred frequency range of the derived filters. Our tests indicate a consistent advantage in phone classification. Word recognition tests, in contrast, reveal that the LDA filters often do not improve upon the existing filters previously used. They can also be made less effectual by allowing contextual frames to a trained probability estimator.