Speech recognition with a new DSP-multiband loudness correction hearing aid algorithm.

Abstract

A new algorithm for a multiband loudness correction hearing aid was implemented in real-time on a digital signal processor (DSP) (TMS320C25, Texas Instruments). Selective amplification of the short-time amplitude spectrum is used to compensate individual hearing losses. The fitting procedure is based on the estimation of loudness growth functions for narrow band signals via a scaling procedure. Frequency and intensity dependent gain factors are calculated for every hearing impaired subject to yield similar loudness growth functions as average normal hearing listeners. To prevent loudness summation across frequency bands while processing broadband signals such as speech, a simplified psychoacoustic loudness (SPL) perception model was incorporated to estimate the loudness of the incoming signal and correct the gain tables accordingly. Speech tests in quiet (at 70 and 60 dB SPL) and noise [at 0 and −5 dB signal-to-noise ratio (SNR)] were performed with 11 hearing aid users. Significant improvements of recognition scores were obtained with the new DSP compared to the subjects own conventional hearing aids [mean improvements in consonant tests: 12% (70 dB), 29% (60 dB), 14% (SNR=0 dB), 12% (SNR=−5 dB); vowel tests: 6% (70 dB), 13% (60 dB), 6% (SNR=0 dB), 11% (SNR=−5 dB); all t-tests significant at p<0.025]. [Work supported by the Swiss National Research Foundation.]