Many audio applications rely on filter banks (FBs) to analyze, process, and re-synthesize sounds. For these applications, an important property of the analysis–synthesis system is the reconstruction error; it has to be minimized to avoid audible artifacts. Other advantageous properties include stability and low redundancy. To exploit some aspects of auditory perception in the signal chain, some applications rely on FBs that approximate the frequency analysis performed in the auditory periphery, the gammatone FB being a popular example. However, current gammatone FBs only allow partial reconstruction and stability at high redundancies. In this article, we construct an analysis–synthesis system for audio applications. The proposed system, referred to as Audlet, is an oversampled FB with filters distributed on auditory frequency scales. It allows perfect reconstruction for a wide range of FB settings (e.g., the shape and density of filters), efficient FB design, and adaptable redundancy. In particular, we show how to construct a gammatone FB with perfect reconstruction. Experiments demonstrate performance improvements of the proposed gammatone FB when compared to current gammatone FBs in terms of reconstruction error and stability, especially at low redundancies. An application of the framework to audio source separation illustrates its utility for audio processing.

This paper describes a method for obtaining a perceptually motivated and perfectly invertible time-frequency representation of a sound signal. Based on frame theory and the recent non-stationary Gabor transform, a linear representation with resolution evolving across frequency is formulated and implemented as a non-uniform filterbank. To match the human auditory time-frequency resolution, the transform uses Gaussian windows equidistantly spaced on the psychoacoustic “ERB” frequency scale. Additionally, the transform features adaptable resolution and redundancy. Simulations showed that perfect reconstruction can be achieved using fast iterative methods and preconditioning even using one filter per ERB and a very low redundancy (1.08). Comparison with a linear gammatone filterbank showed that the ERBlet approximates well the auditory time-frequency resolution.

The additivity of nonsimultaneous masking was studied using Gaussian-shaped tone pulses (referred to as Gaussians) as masker and target stimuli. Combinations of up to four temporally separated Gaussian maskers with an equivalent rectangular bandwidth of 600 Hz and an equivalent rectangular duration of 1.7 ms were tested. Each masker was level-adjusted to produce approximately 8 dB of masking. Excess masking (exceeding linear additivity) was generally stronger than reported in the literature for longer maskers and comparable target levels. A model incorporating a compressive input/output function, followed by a linear summation stage, underestimated excess masking when using an input/output function derived from literature data for longer maskers and comparable target levels. The data could be predicted with a more compressive input/output function. Stronger compression may be explained by assuming that the Gaussian stimuli were too short to evoke the medial olivocochlear reflex (MOCR), whereas for longer maskers tested previously the MOCR caused reduced compression. Overall, the interpretation of the data suggests strong basilar membrane compression for very short stimuli. (C) 2011 Acoustical Society of America. [DOI: 10.1121/1.3518781]

Peripheral compression is believed to play a major role in the masker phase effect (MPE). While compression is almost instantaneous, activation of the efferent system reduces compression in a temporally evolving manner. To study the role of efferent-controlled compression in the MPE, in experiment 1, simultaneous masking of a 30-ms 4-kHz tone by 40-ms Schroeder-phase harmonic complexes was measured with on- and off-frequency precursors as a function of masker phase curvature for two masker levels (60 and 90 dB sound pressure level). The MPE was quantified by the threshold range [min/max difference (MMD)] across the phase curvatures. For the 60-dB condition, the presence of on-frequency precursor decreased the MMD from 10 to 5 dB. Experiment 2 studied the role of the precursor on the auditory filter's bandwidth. The on-frequency precursor was found to increase the bandwidth, an effect incorporated in the subsequent modeling. A model of the auditory periphery including cochlear filtering and basilar membrane compression generally underestimated the MMDs. A model based on two-step compression, including compression of inner hair cells, accounted for the MMDs across precursor and level conditions. Overall, the observed precursor effects and the model predictions suggest an important role of compression in the simultaneous MPE.

Laback et al. [(2011). J. Acoust. Soc. Am. 129, 888-897] investigated the additivity of non-simultaneous masking using short Gaussian-shaped tones as maskers and target. The present study involved Gaussian stimuli to measure the additivity of simultaneous masking for combinations of up to four spectrally separated maskers. According to most basilar membrane measurements, the maskers should be processed linearly at the characteristic frequency (CF) of the target. Assuming also compression of the target, all masker combinations should produce excess masking (exceeding linear additivity). The results for a pair of maskers flanking the target indeed showed excess masking. The amount of excess masking could be predicted by a model assuming summation of masker-evoked excitations in intensity units at the target CF and compression of the target, using compressive input/output functions derived from the nonsimultaneous masking study. However, the combinations of lower-frequency maskers showed much less excess masking than predicted by the model. This cannot easily be attributed to factors like off-frequency listening, combination tone perception, or between-masker suppression. It was better predicted, however, by assuming weighted intensity summation of masker excitations. The optimum weights for the lower-frequency maskers were smaller than one, consistent with partial masker compression as indicated by recent psychoacoustic data. (C) 2013 Acoustical Society of America.

This paper describes a method to obtain a perceptually relevant sparse representation of a sound signal. Based on matching pursuit (MP) and recent psychoacoustic data on time-frequency masking measured with Gabor atoms, a perceptual matching pursuit (PMP) algorithm is proposed. To obtain a good match between the masking model and the signal representation, a dictionary of Gabor atoms with variable sizes is chosen for MP. In the proposed method, the signal is first decomposed using MP and the masking model is applied on the resulting set of atoms. This allows for isolating the masked components from the residual. Experimental results show that exploiting time-frequency masking allows to remove more atoms than using only spectral masking. Additionally, accounting for masking effects between atoms of different sizes and at different times allows for sparser representations. The objective evaluation of the proposed PMP algorithm indicates imperceptible distortions.