Principal-components analysis for low-redundancy encoding of speech spectra

Abstract

The principal-components statistical procedure for data reduction is used to efficiently encode speech power spectra by exploiting the correlations of power spectral amplitudes at various frequencies. Although this datareduction procedure has been used in several previous studies, little attempt was made to optimize the methods for spectral selection and coding through the use of intelligibility testing. In the present study, principal-components basis vectors were computed from the continuous speech of several male and female speakers using various nonlinear spectral amplitude scales. Speech was synthesized using a combination linear predictive (LP) principal-components vocoder. Of the amplitude scales investigated for use with a principalcomponents analysis of speech spectra, logarithmic amplitude coding of non-normalized spectra emerged as a slight favorite. Speech synthesized from four principal components was found to be about 80% intellig/•ble using a form of the Diagnostic Rhyme Test for rhyming word pairs and about 95% intelligible, for words within a sentence context. Speech synthesized from spectral principal components compared favorably in intelligibility and quality with speech synthesized from a control LP vocoder with the same number of parameters.

Some Perceptual Aspects of Timbre

“…Thus this analysis provides a solid experimental basis for construction of a vowel circle which is topologically very similar to that proposed by Yilmaz. Following the lead of Pols and Taylor, we have applied principal components analysis to our Watchdog sentence, Figure 9, in an attempt to generate a second neural representation which is more compact than the one discussed in the first section. The basis vectors generated by this analysis turned out to be very similar in form to those obtained in other studies on completely different material (see Pols 1977& Zahorian 1978. When our basis vectors are used to transform the data of Figure 9, we obtain the new representation shown in Figure 11.…”

One factor underlies individual differences in auditory informational masking within and across age groups

“…Without focusing on specific detection strategies, we tested the possibility of one or more detection strategies by using a principal components analysis (PCA) to evaluate whether the variance in the masking functions arises from one or more underlying sources. PCA is a statistical analysis that is often used in cases where one wishes to represent the underlying structure of a highly variable set of correlated measures (for specific applications in psychoacoustics see Kistler & Wightman, 1992;Pols, van der Kamp, & Plomp, 1969;Zahorian & Rothenberg, 1981). The goal is to represent measures (individual masking functions in the present case) as a weighted sum of a small number of underlying basis functions, each representing an independent source of variance contributing to the total variance among measures.…”

Proper Orthogonal Decomposition (Pod) of a Class of Vibroimpact Oscillations

“…Equation (2) means that the sum of the di!erences squared between v L (x) and (x) should be minimized. In other words, the ensemble average of the inner products between v L and must be maximized.…”

“…In acoustical and random signal decomposition, the Karhunen}Loeve decomposition method (also known as proper orthogonal decomposition) has been widely used to ascertain the modes and the energy of the signals under consideration [1,2]. This is very important in applications that involve compression and storage of stochastic signals.…”