We propose a model that predicts harmonic phases at glottal closure instants. Phases are obtained from the scaled harmonic amplitude envelope derivative. This method is able to generate convincing synthesis results while avoids typical phasiness artifacts. A clear advantage of such model is to simplify the sample concatenation of sample based synthesizers. In addition, it helps to improve the sound quality of voice transformations in several contexts.

Download

We propose a strategy for integrating descriptor-driven transformation into mosaicing sound synthesis, in which samples are selected by taking into account potential distances in the transformed space. Target descriptors consisting of chroma, mel-spaced filter banks, and energy are modeled with respect to windowed bandlimited resampling and mel-spaced filters, and later corrected with gain. These transformations, however simple, allow some adaptation of textural sound material to musical contexts.

Download

In this paper we evaluate some of the alternative methods commonly applied in the first stages of the signal processing chain of automatic melody extraction systems. Namely, the first two stages are studied – the extraction of sinusoidal components and the computation of a time-pitch salience function, with the goal of determining the benefits and caveats of each approach under the specific context of predominant melody estimation. The approaches are evaluated on a data-set of polyphonic music containing several musical genres with different singing/playing styles, using metrics specifically designed for measuring the usefulness of each step for melody extraction. The results suggest that equal loudness filtering and frequency/amplitude correction methods provide significant improvements, whilst using a multi-resolution spectral transform results in only a marginal improvement compared to the standard STFT. The effect of key parameters in the computation of the salience function is also studied and discussed.

Download

This paper describes an improvement of the generalized reassignment method for estimating the parameters of a modulated real sinusoid. The main disadvantage of this method is decreased accuracy for high log-amplitude and/or frequency changes. One of the reasons for such accuracy deterioration stems from the use of the Fourier transform. Fourier transform belongs to a more general family of integral transforms and can be defined as an integral transform using a Fourier kernel function - a stationary complex sinusoid. A correlation between the Fourier kernel function and a non-stationary sinusoid decreases as the modulation of the sinusoid increases, ultimately causing the parameter estimation deterioration. In this paper, the generalized reassignment is reformulated for use with an arbitrary kernel. Specifically, an adaptive polynomial-phase Fourier kernel is proposed. It is shown that such an algorithm needs the parameter estimates from the original generalized reassignment method and that it improves the Signalto-Residual ratio (SRR) in the non-noisy cases. The drawbacks concerning the initial conditions and ways of avoiding a close-tosingular system of linear equations are discussed.

Download

The most common sinusoidal models for non-stationary analysis represent either complex amplitude modulated exponentials with exponential damping (cPACED) or log-amplitude/frequency modulated exponentials (generalised sinusoids), by far the most commonly used modulation function being polynomials for both signal families. Attempts to tackle a hybrid sinusoidal model, i.e. a generalised sinusoid with complex amplitude modulation were relying on approximations and iterative improvement due to absence of a tractable analytical expression for their Fourier Transform. In this work a simple, direct solution for the aforementioned model is presented.

Download