Showing papers by "Ramdas Kumaresan published in 1995"

Dynamic tracking filters for decomposing nonstationary sinusoidal signals

Abstract: A procedure to decompose a signal consisting of nonstationary sinusoidal components based on the principles of residual signal analysis is proposed. A tracking unit consisting of an all zero filter (AZF) in cascade with a dynamic tracking filter (DTF) is assigned to each component. While the adaptively varying zeros of the AZF suppresses all interfering neighbors, the DTF captures the slowly varying instantaneous frequency (IF) of the desired component. Our earlier methods improved upon Costas's estimator-predictor filter bank by using a better interfering signal predictor. The alternative procedure described in this paper avoids the use of prediction and is not overly restricted by the number of components. We also show that by using two simple feedback loops (a loop-filter is thus avoided) the tracking information is ensured to be in phase. Finally, the algorithms ability to decompose synthetic signals, speech signals into harmonic partials as well as tracking the formants present in voiced speech segments is illustrated.

Minimum/maximum phase decomposition of signals inspired by the auditory periphery

Abstract: We propose a signal decomposition approach based on our interpretation of the functioning of the auditory periphery. Analogous to previously proposed auditory models, a parallel-bank of M filters forms the front-end of our model. However, each filter output is further decomposed into a minimum-phase (MinP) and a maximum-phase (MaxP) component. These components are analytic signals with special envelope-phase relationships. Once the filter output is decomposed into MinP/MaxP components we compute the instantaneous frequencies of these components. Finally, we sift through the instantaneous frequencies of all the MinP components and retain only those filter outputs that are entirely MaxP. It appears that a handful of instantaneous frequency tracks of MaxP components are sufficient to represent vowel sounds.

Effect of model mismatch on polynomial envelope and phase modeling of nonstationary sinusoids

Abstract: In this paper we examine the effect of model mismatch when modeling a signal consisting of multiple non-stationary sinusoids. The envelopes and frequencies of the components were modeled as polynomials of low order over short intervals of time and the coefficients estimated using the least-squares error criterion. If the block length and model orders are not properly chosen, unacceptable errors occurred in the estimated frequency tracks. The errors tended to increase as the number of components increased. Using the simpler constant envelope, constant frequency sinewave model for short, heavily overlapping blocks and smoothing the resulting frequency tracks gave surprisingly good results when analyzing a long duration multicomponent signal. We conclude that the more complex polynomial model may not always yield the expected increase in accuracy in signal modeling.