Miroslav Zivanovic

Measurement of vibrato in lyric singers

In this paper, some digital signal processing techniques are proposed to measure the properties of the vibrato performed by lyric singers. From an audio record, three main signals are obtained which correlate with the acoustically perceived parameters: intonation, rate, and extent. The results obtained are much more informative than those previously reported by other authors who use less sophisticated techniques.

Adaptive threshold determination for spectral peak classification

A new approach to adaptive threshold selection for classification of peaks of audio spectra is presented. We here extend the previous work on classification of sinusoidal and noise peaks based on a set of spectral peak descriptors in a twofold way: on one hand we propose a compact sinusoidal model where all the modulation parameters are defined with respect to the analysis window. This fact is of great importance as we recall that the STFT spectra are closely related to the analysis window properties. On the other hand, we design a threshold selection algorithm that allows us to control the decision thresholds in an intuitive manner. The decision thresholds calculated from the relationships established between the noise power in the signal and the distributions of sinusoidal peaks assures that all peaks described as sinusoidal will be correctly classified. We also show that the threshold selection algorithm can be used for different types of analysis windows with only a slight parameter readjustment.

Simultaneous powerline interference and baseline wander removal from ECG and EMG signals by sinusoidal modeling.

We present a compact approach to joint modeling of powerline interference (PLI) and baseline wonder (BW) for denoising of biopotential signals. Both PLI and BW are modeled by a set of harmonically related sinusoids modulated by low-order time polynomials. The sinusoids account on the harmonicity and mean instantaneous frequency of the PLI in the analysis window, while the polynomials capture the frequency and amplitude deviations from their nominal values and characterize the BW at the same time. The resulting model is linear-in-parameters and the solution to the corresponding linear system is estimated in a simple and efficient way through linear least-squares. The proposed modeling method was evaluated on real electrocardiographic (ECG) and electromyographic (EMG) signals against three reference methods for different analysis scenarios. The comparative study suggests that the proposed method outperforms the reference methods in terms of residual interference energy in the denoised biopotential signals.

Instrument for the measurement of the instantaneous frequency

This paper describes the design and implementation of an instrument for the measurement of the instantaneous frequency of a signal. It is based on the extensive use of signal processing techniques and their implementation on efficient processors (digital signal processors). The concept of instantaneous frequency is first reviewed and some procedures for its estimation are evaluated with an off-line virtual instrument. The most promising techniques were then implemented on a PC-based instrument, achieving a real-time operation for audio ranges.

Nonparametric spectrum interpolation methods: a comparative study

In this paper, a number of nonparametric methods that allow a modification in the computational resolution of the discrete spectrum are reviewed and compared according to different criteria. In this way, the most adequate method can be selected for a given application, according to the guidelines given in the paper. Some novel results are given in the analysis of the compared methods, in particular, an approximate method for the calculation of the signal-to-noise ratio using the frequency warping technique.

Instantaneous frequency and amplitude of vibrato in singing voice

We investigate the relationship between the instantaneous amplitude (IA) and the instantaneous frequency (IF) of vibrato signals in the singing voice. It is shown that this relationship is of great value in obtaining information about the vocal tract model. However, to make this analysis possible, it is necessary to cope with two basic limitations: reverberation in recordings, which shows up as multiharmonicity in each partial, and the phase effect of vocal tract formants which distort the instantaneous frequency of some partials.

On The Polynomial Approximation for Time-Variant Harmonic Signal Modeling

We present a novel approach to modeling time-variant harmonic content in monophonic audio signals. We show that both amplitude and fundamental frequency time variations can be compactly captured in a single time polynomial which modulates the fundamental harmonic component. A correct estimation of the fundamental frequency is assured through the fully automated spectral analysis method (ASA). The best-fit is easily obtained by linear least-squares, given the fact that the set of equations is linear-in-parameters. In contrast to the existing methods, the proposed approach is designed to properly describe harmonic structures in monophonic audio signals under conditions of both amplitude and frequency variations and low signal-to noise ratios.

Extending the limits of resolution for narrow-band harmonic and modal analysis: a non-parametric approach

Non-parametric fast Fourier transform-based spectral analysis is an efficient tool for characterization of signals and systems in the frequency domain, but it often suffers from insufficient spectral resolution. Here we propose methods for improving the physical resolution of spectra by redistributing the original spectral information to emulate larger-time observations. This is accomplished either by non-parametric destructive discrete Fourier transform interpolation or by asymmetric time-windowing. The proposed methods are shown to have very good peak discrimination capacity, which together with their low computational complexity makes them good candidates for narrow-band spectral analysis in instrumentation.

Wavelet-based unsupervised learning method for electrocardiogram suppression in surface electromyograms.

We present a novel approach aimed at removing electrocardiogram (ECG) perturbation from single-channel surface electromyogram (EMG) recordings by means of unsupervised learning of wavelet-based intensity images. The general idea is to combine the suitability of certain wavelet decomposition bases which provide sparse electrocardiogram time-frequency representations, with the capacity of non-negative matrix factorization (NMF) for extracting patterns from images. In order to overcome convergence problems which often arise in NMF-related applications, we design a novel robust initialization strategy which ensures proper signal decomposition in a wide range of ECG contamination levels. Moreover, the method can be readily used because no a priori knowledge or parameter adjustment is needed. The proposed method was evaluated on real surface EMG signals against two state-of-the-art unsupervised learning algorithms and a singular spectrum analysis based method. The results, expressed in terms of high-to-low energy ratio, normalized median frequency, spectral power difference and normalized average rectified value, suggest that the proposed method enables better ECG-EMG separation quality than the reference methods.

Single and Piecewise Polynomials for Modeling of Pitched Sounds

We present a compact approach to simultaneous modeling of non-stationary harmonic and transient components in pitched sound sources. The harmonic and transient components are described by separate models which are built from a common sinusoidal basis modified by a joint action of single and linear piecewise time polynomials respectively. A single polynomial accounts for slow and continuous signal time variations, while various piecewise polynomials can capture fast signal changes on smaller subintervals within the analysis window. The resulting model is linear-in-parameters and the solution to the corresponding linear system of equations provides correct model parameter estimates according to the signal content in the analysis window. The model is extended to deal with mixtures of sounds, where harmonics clustered in a small bandwidth are jointly modeled as a single harmonic. The comparative results suggest that the proposed model outperforms two reference modeling methods in terms of modeling errors and number of parameters.