Gain-induced speech distortions and the absence of intelligibility benefit with existing noise-reduction algorithms

Abstract

Most noise-reduction algorithms used in hearing aids apply a gain to the noisy envelopes to reduce noise interference. The present study assesses the impact of two types of speech distortion introduced by noise-suppressive gain functions: amplification distortion occurring when the amplitude of the target signal is over-estimated, and attenuation distortion occurring when the target amplitude is under-estimated. Sentences corrupted by steady noise and competing talker were processed through a noise-reduction algorithm and synthesized to contain either amplification distortion, attenuation distortion or both. The attenuation distortion was found to have a minimal effect on speech intelligibility. In fact, substantial improvements (> 80 percentage points) in intelligibility, relative to noise-corrupted speech, were obtained when the processed sentences contained only attenuation distortion. When the amplification distortion was limited to be smaller than 6 dB, performance was nearly unaffected in the steady-noise conditions, but was severely degraded in the competing-talker conditions. Overall, the present data suggest that one reason that existing algorithms do not improve speech intelligibility is because they allow amplification distortions in excess of 6 dB. These distortions are shown in this study to be always associated with masker-dominated envelopes and should thus be eliminated.

Comparison of two channel selection criteria for noise suppression in cochlear implants

“…Further analysis of the data obtained with the five Nucleus subjects tested indicated that performance obtained with the maximum-selection criterion (ACE) and CIS was significantly (p < 0.003) lower than performance obtained using the SNR-and C-rule (T 0 ¼ 1, T 0 ¼ 2) channel selection criteria for all input SNRs. These outcomes are consistent with previous findings with CI users (Hu and Loizou, 2008) and NH listeners (Brungart et al, 2006;Li and Loizou, 2008;Kim and Loizou, 2011).…”

“…While not tested here, large intelligibility benefits can be obtained using channelselection based coding strategies in other masker conditions including competing talker and multi-talker babble. Prior tests with NH listeners, for instance, indicated no dependency of intelligibility benefit on the type of masker used (Li and Loizou, 2008;Kim and Loizou, 2011) even when the noise spectrum was not known a priori and was estimated from the noisy signal (Kim and Loizou, 2010). Second, it is not necessary to increase the number of effective channels of information transmitted by CIs to improve speech intelligibility in noise (e.g., Friesen et al, 2001).…”

“…The preceding channel-selection criterion cannot be easily incorporated in noise suppression algorithms as it does not make use of any information contained in the noisesuppressed output envelopes. For that reason, a different channel selection criterion was developed Kim and Loizou, 2011) and found to yield substantial improvements in speech intelligibility by NH listeners. The new channel selection criterion is based on imposing constraints on the noise-suppressed envelopes, and we thus refer to it as the constrained channel-selection rule and denote it as the C-rule.…”

Modifying the normalized covariance metric measure to account for nonlinear distortions introduced by noise-reduction algorithms

“…The shape of the gain function varies across algorithms, but independent of its shape, when the gain function is applied to the mixture envelopes (or spectra), it introduces nonlinear distortions to the speech envelopes or spectra (i.e., the difference between the noise-reduced speech and the clean speech). Earlier studies (e.g., Loizou and Ma, 2011;Kim and Loizou, 2011;G omez et al, 2012) showed that it is beneficial to account for the effects of these nonlinear distortions in the objective intelligibility measures.…”

“…In an earlier effort to address this issue, Kim and Loizou (2011) simplified the classification of nonlinear distortions and classified the distortions into either amplification or attenuation distortions. The amplification distortion refers to the scenario that the envelope or magnitude spectrum of the noise-suppressed speech is larger than that of the clean speech; and the attenuation distortion refers to the scenario that the envelope or magnitude spectrum of the noisesuppressed speech is smaller than that of the clean speech.…”