Ronald A. J. van Elburg

Coen Jonker, Arryon D. Tijsma, Ronald A. J. van Elburg

Time-frequency representations are important for the analysis of time series. We have developed an online time-series analysis system and equipped it to reliably handle re-alignment in the time-frequency plane. The system can deal with issues like invalid regions in time-frequency representations and discontinuities in data transmissions, making it suitable for on-line processing in real-world situations. In retrospect the whole problem can be considered to be a generalization of ideas present in overlap-and-add filtering, but then for time-frequency representations and including the calculation of non-causal features. Here we present our design for time-frequency representation fission and fusion rules. We present these rules in the context of two typical use cases, which facilitate understanding of the underlying choices.

Ronald A. J. van Elburg, Tjeerd C. Andringa

Human categorization of sound seems predominantly based on sound source properties. To estimate these source properties we propose a novel sound analysis method, which separates sound into different sonic textures: tones, pulses, and broadband noises. The audible presence of tones or pulses corresponds to more extended cochleagram patterns than can be expected on the basis of correlations introduced by the gammachirp filterbank alone. We design tract features to respond to these extended patterns, and use these to identify areas of the time-frequency plane as tonal, pulsal, and noisy. Where an area is marked as noisy if it is neither tonal nor pulsal. To investigate whether a similar separation indeed underlies human perceptual organization we introduce tract based descriptors: tonality, pulsality, and noisiness. These descriptors keep track of either the total energy or the cochleagram area marked as respectively tonal, pulsal, and noisy. Energy based tonality and pulsality is strongly correlated with the first perceptual dimension of human subjects, while energy based noisiness correlates moderately with the second perceptual dimension. We conclude that harmonic, impact and continuous process sounds can be largely separated with energy based tonality, pulsality and noisiness.