Speech recognition with a hearing-aid processing scheme combining beamforming with mask-informed speech enhancement.

A signal processing approach combining beamforming with mask-informed speech enhancement was assessed by measuring sentence recognition in listeners with mild-to-moderate hearing impairment in adverse listening conditions that simulated the output of behind-the-ear hearing aids in a noisy classroom. Two types of beamforming were compared: binaural, with the two microphones of each aid treated as a single array, and bilateral, where independent left and right beamformers were derived. Binaural beamforming produces a narrower beam, maximising improvement in signal-to-noise ratio (SNR), but eliminates the spatial diversity that is preserved in bilateral beamforming. Each beamformer type was optimised for the true target position and implemented with and without additional speech enhancement in which spectral features extracted from the beamformer output were passed to a deep neural network trained to identify time-frequency regions dominated by target speech. Additional conditions comprising binaural beamforming combined with speech enhancement implemented using Wiener filtering or modulation-domain Kalman filtering were tested in normally-hearing (NH) listeners. Both beamformer types gave substantial improvements relative to no processing, with significantly greater benefit for binaural beamforming. Performance with additional mask-informed enhancement was poorer than with beamforming alone, for both beamformer types and both listener groups. In NH listeners the addition of mask-informed enhancement produced significantly poorer performance than both other forms of enhancement, neither of which differed from the beamformer alone. In summary, the additional improvement in SNR provided by binaural beamforming appeared to outweigh loss of spatial information, while speech understanding was not further improved by the mask-informed enhancement method implemented here.

Forward Masking in Cochlear Implant Users: Electrophysiological and Psychophysical Data Using Pulse Train Maskers.

Electrical stimulation of auditory nerve fibers using cochlear implants (CI) shows psychophysical forward masking (pFM) up to several hundreds of milliseconds. By contrast, recovery of electrically evoked compound action potentials (eCAPs) from forward masking (eFM) was shown to be more rapid, with time constants no greater than a few milliseconds. These discrepancies suggested two main contributors to pFM: a rapid-recovery process due to refractory properties of the auditory nerve and a slow-recovery process arising from more central structures. In the present study, we investigate whether the use of different maskers between eCAP and psychophysical measures, specifically single-pulse versus pulse train maskers, may have been a source of confound.In experiment 1, we measured eFM using the following: a single-pulse masker, a 300-ms low-rate pulse train masker (LTM, 250 pps), and a 300-ms high-rate pulse train masker (HTM, 5000 pps). The maskers were presented either at same physical current (Φ) or at same perceptual (Ψ) level corresponding to comfortable loudness. Responses to a single-pulse probe were measured for masker-probe intervals ranging from 1 to 512 ms. Recovery from masking was much slower for pulse trains than for the single-pulse masker. When presented at Φ level, HTM produced more and longer-lasting masking than LTM. However, results were inconsistent when LTM and HTM were compared at Ψ level. In experiment 2, masked detection thresholds of single-pulse probes were measured using the same pulse train masker conditions. In line with our eFM findings, masked thresholds for HTM were higher than those for LTM at Φ level. However, the opposite result was found when the pulse trains were presented at Ψ level.Our results confirm the presence of slow-recovery phenomena at the level of the auditory nerve in CI users, as previously shown in animal studies. Inconsistencies between eFM and pFM results, despite using the same masking conditions, further underline the importance of comparing electrophysiological and psychophysical measures with identical stimulation paradigms.

Pulse-spreading harmonic complex as an alternative carrier for vocoder simulations of cochlear implants.

Noise- and sine-carrier vocoders are often used to acoustically simulate the information transmitted by a cochlear implant (CI). However, sine-waves fail to mimic the broad spread of excitation produced by a CI and noise-bands contain intrinsic modulations that are absent in CIs. The present study proposes pulse-spreading harmonic complexes (PSHCs) as an alternative acoustic carrier in vocoders. Sentence-in-noise recognition was measured in 12 normal-hearing subjects for noise-, sine-, and PSHC-vocoders. Consistent with the amount of intrinsic modulations present in each vocoder condition, the average speech reception threshold obtained with the PSHC-vocoder was higher than with sine-vocoding but lower than with noise-vocoding.

Optimizing pulse-spreading harmonic complexes to minimize intrinsic modulations after auditory filtering.

All signals, except sine waves, exhibit intrinsic modulations that affect perceptual masking. Reducing the physical intrinsic modulations of a broadband signal does not necessarily have a perceptual impact: auditory filtering can reintroduce modulations. Broadband signals with low intrinsic modulations after auditory filtering have proved difficult to design. To that end, this paper introduces a class of signals termed pulse-spreading harmonic complexes (PSHCs). PSHCs are generated by summing harmonically related components with such a phase that the resulting waveform exhibits pulses equally-spaced within a repetition period. The order of a PSHC determines its pulse rate. Simulations with a gamma-tone filterbank suggest an optimal pulse rate at which, after auditory filtering, the PSHC's intrinsic modulations are lowest. These intrinsic modulations appear to be less than those for broadband pseudo-random (PR) or low-noise (LN) noise. This hypothesis was tested in a modulation-detection experiment involving five modulation rates ranging from 8 to 128 Hz and both broadband and narrowband carriers using PSHCs, PR, and LN noise. PSHC showed the lowest thresholds of all broadband signals. Results imply that optimized PSHCs exhibit less intrinsic modulations after auditory filtering than any other broadband signal previously considered.

Effects of noise suppression on intelligibility. II: An attempt to validate physical metrics.

Using the data presented in the accompanying paper [Hilkhuysen et al., J. Acoust. Soc. Am. 131, 531-539 (2012)], the ability of six metrics to predict intelligibility of speech in noise before and after noise suppression was studied. The metrics considered were the Speech Intelligibility Index (SII), the fractional Articulation Index (fAI), the coherence intelligibility index based on the mid-levels in speech (CSIImid), an extension of the Normalized Coherence Metric (NCM+), a part of the speech-based envelope power model (pre-sEPSM), and the Short Term Objective Intelligibility measure (STOI). Three of the measures, SII, CSIImid, and NCM+, overpredicted intelligibility after noise reduction, whereas fAI underpredicted these intelligibilities. The pre-sEPSM metric worked well for speech in babble but failed with car noise. STOI gave the best predictions, but overall the size of intelligibility prediction errors were greater than the change in intelligibility caused by noise suppression. Suggestions for improvements of the metrics are discussed.

Effects of noise suppression on intelligibility: experts' opinions and naive normal-hearing listeners' performance.

PURPOSE: In this study, the authors investigated how well experts can adjust the settings of a commercial noise-reduction system to optimize the intelligibility for naive normal-hearing listeners. METHOD: In Experiment 1, 5 experts adjusted parameters for a noise-reduction system while aiming to optimize intelligibility. The stimuli consisted of speech presented in car-cabin noise or babble at 5 different signal-to-noise ratios (SNRs). In Experiment 2, the effects of processing with these settings were measured with 10 listeners undertaking an intelligibility test. In Experiment 3, the intelligibility of a broad range of settings was investigated with another 10 listeners to determine whether the experts' chosen settings could have been improved. RESULTS: Low Cronbach's alphas indicated that parameter settings varied considerably within and across experts. For very low SNRs, mean proposed settings differed from those for higher SNRs. The different settings had no significant effects on intelligibility for naive normal-hearing listeners. At high SNRs, the settings proposed by experts were found to deteriorate intelligibility. Superior intelligibility for naive normal-hearing listeners was achievable from settings other than the ones proposed by the experts. CONCLUSION: While attempting to enhance noisy speech, experts may propose settings that deteriorate intelligibility for naive normal-hearing listeners.

Effects of noise suppression on intelligibility: dependency on signal-to-noise ratios.

The effects on speech intelligibility of three different noise reduction algorithms (spectral subtraction, minimal mean squared error spectral estimation, and subspace analysis) were evaluated in two types of noise (car and babble) over a 12 dB range of signal-to-noise ratios (SNRs). Results from these listening experiments showed that most algorithms deteriorated intelligibility scores. Modeling of the results with a logit-shaped psychometric function showed that the degradation in intelligibility scores was largely congruent with a constant shift in SNR, although some additional degradation was observed at two SNRs, suggesting a limited interaction between the effects of noise suppression and SNR.