A nonlinear squeezing of the continuous wavelet transform based on auditory nerve models. (English) Zbl 0848.92003

Daubechies, Ingrid; Maes, Stéphane

Aldroubi, Akram (ed.) et al., Wavelets in medicine and biology. Boca Raton, NY: CRC Press (ISBN 978-0-8493-9483-6/hbk; 978-0-367-44859-2/pbk; 978-0-203-73403-2/ebook). 527-546 (1996).

The approach presented in this chapter resulted from a concrete problem in speaker identification. Our goal was to incorporate the wavelet transform and auditory nerve-based models into a tool that could be used for speaker identification (among other applications), in the hope that the results would be more robust to noise than the standard methods.
This chapter is organized as follows. Sections 20.2 to 20.4 present background material, explaining, respectively, (1) how the (continuous) wavelet transform comes up “naturally” in our auditory system; (2) a heuristic approach based on auditory nerve models, which eliminates much of the redundancy in the first-stage transform; and (3) the modulation model, valid for large portions of (voiced) speech, and which is used for speaker identification. In Section 20.5 we put all this background material to use in our own synthesis, an approach that we call “squeezing” the wavelet transform; with an extra refinement this becomes “synchrosqueezing.” The main idea is that the wavelet transform itself has “smeared” out different harmonic components, and that we need to “refocus” the resulting time-frequency or time-scale picture. How this is done is explained in Section 20.5. Section 20.6 deals with various implementation issues, which are touched upon rather than explained in detail. Finally, Section 20.7 shows some results: the “untreated” wavelet transform of a speech segment, its squeezed and synchrosqueezed versions, and the extraction of the parameters used for speaker identification. We conclude with some pointers to and comparisons with similar work in the literature, and with sketching possible future directions.
For the entire collection see [Zbl 0840.00056].

MSC:

92C20	Neural biology
92C99	Physiological, cellular and medical topics

Keywords:

speaker identification; wavelet transform; auditory nerve-based models; modulation model