This paper presents an index designed to predict music quality for individuals listening through hearing aids. The index is "intrusive", that is, it compares the degraded signal being evaluated to a reference signal. The index is based on a model of the auditory periphery that includes the effects of hearing loss. Outputs from the auditory model are used to measure changes in the signal time-frequency envelope modulation, temporal fine structure, and long-term spectrum caused by the hearing aid processing. The index is constructed by combining a term sensitive to noise and nonlinear distortion with a second term sensitive to changes in the long-term spectrum. The index is fitted to an existing database of music quality judgments made by listeners having normal or impaired hearing. The data comprise ratings for three music excerpts (classical orchestra, jazz trio, and jazz singer), each processed through 100 conditions representative of hearing-aid processing and listening situations. The overall accuracy of the index is high, with a correlation coefficient of 0.970 when computed over all of the processing conditions and averaged over the combined groups of listeners having normal and impaired hearing.

Speech quality metrics are used to predict quality under a variety of noise and distortion conditions. A simple metric for noisy speech is the signal-to-noise ratio (SNR), which gives the ratio of the long-term average speech power to the long-term average noise power. However, the average noise level fails to include the impact of the noise amplitude modulation on speech quality judgments. This paper reviews how the noise characteristics affect quality ratings even when the SNR is held constant. In one study, ratings for speech combined with stationary Gaussian noise are compared to ratings for noise having the same envelope modulation as the speech. In a second study, ratings for speech combined with continuous noise or multi-talker babble are compared to speech where the interference is gated off during the silent intervals in the stimulus. An accurate quality metric must take into account the modulation as well as the intensity of the noise, and the implications for speech quality metrics are discussed.

In this paper, the family of conditional minimum mean square error (MMSE) spectral estimators is studied which take on the form (E(XEalphap/|Xp+ Dp|))1alpha/, where Xpis the clean speech spectrum, and Dpis the noise spectrum, resulting in a generalized MMSE estimator (GMMSE). The degree of noise suppression versus musical tone artifacts of these estimators is studied. The tradeoffs in selection of (alpha), across noise spectral structure and signal-to-noise ratio (SNR) level, are also considered. Members of this family of estimators include the Ephraim-Malah (EM) amplitude estimator and, for high SNRs, the Wiener Filter. It is shown that the colorless residual noise observed in the EM estimator is a characteristic of this general family of estimators. An application of these estimators in an auditory enhancement scheme using the masking threshold of the human auditory system is formulated, resulting in the GMMSE-auditory masking threshold (AMT) enhancement method. Finally, a detailed evaluation of the proposed algorithms is performed over the phonetically balanced TIMIT database and the National Gallery of the Spoken Word (NGSW) audio archive using subjective and objective speech quality measures. Results show that the proposed GMMSE-AMT outperforms MMSE and log-MMSE enhancement methods using a detailed phoneme-based objective quality analysis

While a number of studies have investigated various speech enhancement and noise suppression schemes, most consider either a single channel or array processing framework. Clearly there are potential advantages in leveraging the strengths of array processing solutions in suppressing noise from a direction other than the speaker, with that seen in single channel methods that include speech spectral constraints or psychoacoustically motivated processing. In this paper, we propose to integrate a combined fixed/adaptive beamforming algorithm (CFA-BF) for speech enhancement with two single channel methods based on speech spectral constrained iterative processing (Auto-LSP), and an auditory masked threshold based method using equivalent rectangular bandwidth filtering (GMMSE-AMTERB). After formulating the method, we evaluate performance on a subset of the TIMIT corpus with four real noise sources. We demonstrate a consistent level of noise suppression and voice communication quality improvement using the proposed method as reflected by an overall average 26dB increase in SegSNR from the original degraded audio corpus.