Search

A computationally efficient system for sound environment classification in digital hearing aids is presented in this paper. The goal is to automatically classify three different listening environments: "speech," "music," and "noise." The system is designed considering the computational limitations found in such devices. The proposed algorithm is based on a novel set of heuristically designed features inspired in the Mel frequency cepstral coefficients. Experiments carried out with real signals demonstrate that the three listening environments can be robustly classified with the proposed system, obtaining low error rates when using a small part of the total computational resources of the DSP of the device. This study demonstrates that the proposed system can be implemented with the available resources in state-of-the-art digital hearing aids.

Abstract: The improvement of speech intelligibility in hearing aids is still an unsolved problem. Modern devices, either monaural or binaural, may include microphone arrays to provide directivity by means of spatial filtering, but the signals that arrive at the array are distorted by the so-called head shadow effect, which must be considered in the design. The fact that these effects are highly dependent on a person causes the design of an array customized for a subject to need a correct measurement of these effects, which is not practical in real scenarios. The lack of information about the head causes directivity reduction and distortions. In this work we propose three different methods to compute optimized filter coefficients with the purpose of maximizing the array gain and minimizing distortions, and with the objective that these filter coefficients are valid for all users. Furthermore, measurements of the amplitude and phase distortions caused by a lack of head information in the design are introduced. In addition, the methods proposed here have been evaluated in 13 different array configurations. Results show that these approaches reduce significantly the gain reduction and distortions, obtaining a similar intelligibility in comparison to the case where head information is available. [Copyright &y& Elsevier]