Finn T. Agerkvist

Improving the efficiency of deconvolution algorithms for sound source localization

The localization of sound sources with delay-and-sum (DAS) beamforming is limited by a poor spatial resolution-particularly at low frequencies. Various methods based on deconvolution are examined to improve the resolution of the beamforming map, which can be modeled by a convolution of the unknown acoustic source distribution and the beamformers response to a point source, i.e., point-spread function. A significant limitation of deconvolution is, however, an additional computational effort compared to beamforming. In this paper, computationally efficient deconvolution algorithms are examined with computer simulations and experimental data. Specifically, the deconvolution problem is solved with a fast gradient projection method called Fast Iterative Shrikage-Thresholding Algorithm (FISTA), and compared with a Fourier-based non-negative least squares algorithm. The results indicate that FISTA tends to provide an improved spatial resolution and is up to 30% faster and more robust to noise. In the spirit of reproducible research, the source code is available online.

Extracting the invariant model from the feedback paths of digital hearing aids

Feedback whistling is a severe problem with hearing aids. A typical acoustical feedback path represents a wave propagation path from the receiver to the microphone and includes many complicated effects among which some are invariant or nearly invariant for all users and in all acoustical environments given a specific type of hearing aids. Based on this observation, a feedback path model that consists of an invariant model and a variant model is proposed. A common-acoustical-pole and zero model-based approach and an iterative least-square search-based approach are used to extract the invariant model from a set of impulse responses of the feedback paths. A hybrid approach combining the two methods is also proposed. The general properties of the three methods are studied using artificial datasets, and the methods are cross-validated using the measured feedback paths. The results show that the proposed hybrid method gives the best overall performance, and the extracted invariant model is effective in modeling the feedback path.

Using a reflection model for modeling the dynamic feedback path of digital hearing aids.

Feedback whistling is one of the severe problems with hearing aids, especially in dynamic situations when the users hug, pick up a telephone, etc. This paper investigates the properties of the dynamic feedback paths of digital hearing aids and proposes a model based on a reflection assumption. The model is compared with two existing models: a direct model and an initialization model, using the measured dynamic feedback paths. The comparison shows that the proposed approach is able to model the dynamic feedback paths more efficiently and accurately in terms of mean-square error and maximum stable gain. The method is also extended to dual-microphone hearing aids to assess the possibility of relating the two dynamic feedback paths through the reflection model. However, it is found that in a complicated acoustic environment, the relation between the two feedback paths can be very intricate and difficult to exploit to yield better modeling of the dynamic feedback paths.

Monaural Separation of Dependent Audio Sources Based on a Generalized Wiener Filter

This paper presents a two-stage approach for single- channel separation of dependent audio sources. The proposed algorithm is developed in the Bayesian framework and designed for general audio signals. In the first stage of the algorithm, the joint distribution of discrete Fourier transform (DFT) coefficients of the dependent sources is modeled by complex Gaussian mixture models in the frequency domain from samples of individual sources to capture the properties of the sources and their correlation. During the second stage, the mixture is separated through a generalized Wiener filter, which takes correlation term and local stationarity into account. The performance of the algorithm is tested on real audio signals. The results show that the proposed algorithm works very well when the dependent sources have comparable variances and linear correlation.

Time-frequency analysis with temporal and spectral resolution as the human auditory system

The human perception of sound is a suitable area for the application of a simultaneous time-frequency analysis, since the ear is selective in both domains. A perfect reconstruction filter bank with bandwidths approximating the critical bands is presented. The orthogonality of the filter makes it possible to examine the masking effect with realistic signals. The tree structure of the filter bank makes it difficult to obtain well-attenuated stop-bands. The use of filters of different length solves this problem.<<ETX>>

Reproduction of nearby sources by imposing true interaural differences on a sound field control approach

In anechoic conditions, the Interaural Level Difference (ILD) is the most significant auditory cue to judge the distance to a sound source located within 1 m of the listeners head. This is due to the unique characteristics of a point source in its near field, which result in exceptionally high, distance dependent ILDs. When reproducing the sound field of sources located near the head with line or circular arrays of loudspeakers, the reproduced ILDs are generally lower than expected, due to physical limitations. This study presents an approach that combines a sound field reproduction method, known as Pressure Matching (PM), and a binaural control technique. While PM aims at reproducing the incident sound field, the objective of the binaural control technique is to ensure a correct reproduction of interaural differences. The combination of these two approaches gives rise to the following features: (i) an accurate reproduction of ILDs is achieved at the head positions considered by the method, (ii) the ILD variations in the vicinity of those positions are smoothed, thus lowering the ILD error, and (iii) the true wavefront is preserved. Given the properties of the presented method, intended distance and directional perception is expected.

A new approach for modelling the dynamic feedback path of digital hearing aids

This paper proposes a reflection model for the dynamic feedback path of digital hearing aids and compares it with two existing models: a direct model and an initialization model, based on the measured dynamic feedback paths. The comparison shows that the proposed model is superior to the existing two models in terms of maximum stable gain (MSG). For hearing aids with dual microphones, the possibility of relating the two dynamic feedback paths is also investigated. It is shown that in a complicated acoustic environment, the relation between the two feedback paths can be very intricate and difficult to exploit in modelling the dynamic feedback paths.