Uniform Filter Bank Research Articles

WMMSE design of digital filter banks with specified composite response

A new method for designing uniform and nonuniform digital filter banks with a specified composite response is presented. The composite response of the filter bank can be met either exactly or to within a given tolerance. We focus on filter banks in which the individual filters are finite impulse response (FIR) digital filters of possibly nonequal length, although the new method is applicable even to more general structures as well. The new method minimizes the weighted sum of the mean square errors in the response of the individual filters, subject to the composite response specifications. Sufficient conditions for either real, ness or phase linearity of the optimal individual filters are presented. The new weighted minimum mean square error (WMMSE) design method is interpreted from a statistical viewpoint as a maximization of the harmonic mean of the output signal-to-noise ratio (SNR) of the individual filters. The complexity of the new method is analyzed, and the design process is demonstrated via a design example.

On the effects of varying filter bank parameters on isolated word recognition

The vast majority of commercially available isolated word recognizers use a filter bank analysis as the front end processing for recognition. It is not well understood how the parameters of different filter banks (e.g., number of filters, types of filters, filter spacing, etc.) affect recognizer performance. In this paper we present results of performance evaluation of several types of filter bank analyzers in a speaker trained isolated word recognition test using dialed-up telephone line recordings. We have studied both DFT (discrete Fourier transform) and direct form implementations of the filter banks. We have also considered uniform and nonuniform filter spacings. The results indicate that the best performance (highest word accuracy) is obtained by both a 15-channel uniform filter bank and a 13-channel nonuniform filter bank (with channels spacing along a critical band scale). The performance of a 7-channel critical band filter bank is almost as good as that of the two best filter banks. In comparison to a conventional linear predictive coding (LPC) word recognizer, the performance of the best filter bank recognizers was, on average, several percent worse than that of an eighth-order LPC-based recognizer. A discussion as to why some filter banks performed better than others, and why the LPC-based system did the best, is given in this paper.

Polynomial transform implementation of digital filter banks

This paper investigates the use of polynomial transforms for the implementation of uniform digital bandpass filter banks. The technique is based upon a decomposition of the N bandpass filters into a set of real polyphase filters followed by a DCT (discrete cosine transform) of size N. The DCT is converted into a DFT (discrete Fourier transform) of size N and the polyphase filters are evaluated by DFT's. This procedure yields a two-dimensional DFT which is computed by a polynomial transform and odd DFT's. We show that this technique reduces significantly the number of arithmetic operations when compared to conventional methods, and yields a regular structure in which most of the computations are performed with FFT-type algorithms.

On the effects of varying filter bank parameters on isolated word recognition

The vast majority of commercially available isolated word recognizers use a filter bank analysis as the front end processing for recognition. It is not well understood how the parameters of different filter banks (e.g., number of filters, types of filters, filter spacing, etc.) affect recognizer performance. In this talk we present results of performance evaluation of several types of filter bank analyzer in a speaker trained, isolated word recognition test using dialed‐up telephone line recordings. We have studied both DFT (discrete Fourier transform) and direct form implementations of the filter banks. We have also considered uniform and nonuniform filter spacings. The results indicate that the best performance (highest word accuracy) is obtained by both a 15‐channel uniform filter bank and a 13‐channel nonuniform filter bank (with channels spaced along a critical band scale). In comparison to a conventional linear predictive coding (LPC) word recognizer, the performance of the best filter bank recognizers was, on average, several percent worse than that of an eighth order LPC‐based recognizer.