Factors governing speech reception benefits of adaptive linear filtering for listeners with sensorineural hearing loss.

Adaptive linear filtering can improve effective speech-to-noise ratios by attenuating spectral regions with intense noise components to reduce the noise's spread of masking onto speech in neighboring regions. This mechanism was examined in static listening conditions for seven individuals with sensorineural hearing loss. Subjects were presented with nonsense syllables in an intense octave-band noise centered on 0.5, 1, or 2 kHz. The nonsense syllables were amplified to maximize. the articulation index; the noises were the same for all subjects. The processing consisted of applying frequency-selective attenuation to the speech-plus-noise with the goal of attenuating the frequency region containing the noise by various amounts. Consonant recognition scores and noise masking patterns were collected in all listening conditions. When compared with masking patterns obtained from normal-hearing subjects, all hearing-impaired subjects had higher masked thresholds at frequencies below, within, and above the masker band except for one subject who demonstrated additional masking above the masker only. Frequency-selective attenuation resulted in both increases and decreases in consonant recognition scores. Increases were associated with a release from upward spread of masking. Decreases were associated with applying too much attenuation such that speech energy within the masker band that was audible before processing was partially below threshold after processing. Fletcher's [Speech and Hearing in Communication (Van Nostrand, New York, 1953)] version of articulation theory (without modification) accounted for individual subject differences within the range of variability associated with the consonant recognition test in almost every instance. Hence, primary factors influencing speech reception benefits are characterized by articulation theory. Fletcher's theory appears well-suited to guide the design of control algorithms that will maximize speech recognition for individual listeners.

Estimating articulation scores.

The ability of listeners to estimate articulation scores for lists of nonsense syllables was evaluated. Normal-hearing subjects were presented with lists of from 50 to 60 nonsense syllables that were degraded with various amounts of noise or filtering and were instructed to estimate consonant-correct scores for each condition. To provide a reference for estimating, subjects were shown the accurate orthographic representation of the syllable on a computer monitor to compare with the auditory presentation. The printed version was displayed either simultaneously with the auditory presentation or 500 ms after the offset of the syllable. Estimates were collected on two occasions to examine test-retest reliability, and actual percent-correct scores were obtained to check the accuracy of the estimates. Most subjects overestimated actual scores when the printed representation was provided simultaneously, but estimates were strikingly similar to actual scores when the printed representation was delayed. The delay appeared to prevent the printed representation from favorably biasing the reception of the syllable. The average of two or three estimates gave highly repeatable results for both visual displays. Crossover frequencies derived from the filtered-speech conditions were within the range reported in the literature. This supports the conclusion that subjects based their estimates on the recognition of speech sounds rather than other percepts associated with the speech-in-noise conditions such as loudness of the noise. The estimation procedure permits the collection of articulation scores in much less time than required by traditional test procedures.

Derivation of frequency-gain characteristics for maximizing speech reception in noise.

Hearing aid gain-assignment schemes known as "prescriptions" were not designed for fitting hearing aids that modify their frequency responses to reduce background noise interference. Rather, prescriptions were developed for hearing aids having single, fixed frequency responses and aim to optimize speech reception in relatively quiet environments. Even though prescriptions do not apply to noisy conditions specifically, they embody the trade between maximizing speech audibility and maintaining loudness comfort that is critical to frequency-gain characteristic selection independent of whether noise is present or absent. The articulation index (AI) was used to examine the extent to which prescriptions' deference to loudness comfort causes them to fall short of maximizing speech spectrum audibility, thereby revealing (roughly) the magnitude of the loudness control built into prescriptions. AIs for speech amplified by an AI-maximizing rule (MAX AI) (Rankovic, Freyman, & Zurek, 1992) and according to several prescriptions were calculated as a function of hearing loss degree and configuration for quiet and noisy conditions. In quiet, AIs for prescriptions were similar to one another when presented with the same audiogram but were drastically smaller than MAX AIs, implying that prescriptions limit speech audibility to a large extent to prevent loudness discomfort. In noise, maximizing the AI required frequency-gain characteristics that were substantially different from prescription-assigned characteristics and that were unique to each noise/audiogram combination. A loudness constraint for the MAX AI scheme was developed to account for the gain discrepancy between prescription AIs and MAX AIs observed in the quiet condition, based on the highest comfortable loudness (HCL) equations presented by Cox (1989) in combination with a loudness model (von Paulus & Zwicker, 1972). The MAX AI scheme with the new loudness control was extended to specify frequency-gain characteristics expected to be optimal for several conditions containing noise, and examples are presented.

Potential benefits of adaptive frequency-gain characteristics for speech reception in noise.

Some current single-microphone hearing aids employ techniques for adaptively varying the frequency-gain characteristics in an attempt to improve speech reception in noise. The potential benefit of this strategy depends on the spectral spread of masking and the degree to which it can be reduced by changing the frequency-gain characteristic. In this study these benefits were examined for subjects with normal hearing under static listening conditions. In the unprocessed condition, subjects were presented with nonsense syllables in an octave-band noise centered on 0.5, 1, or 2 kHz. The frequency-gain characteristic was then modified with the goal of reducing the intensity of the frequency region containing the octave-band noise. This processing resulted in increases as large as 60 percentage points in consonant-correct scores with the low- and mid-frequency octave noise bands, and a small increase with the high-frequency noise. Masking patterns produced by the octave noises were also measured and were related to the intelligibility results via an analysis based on Articulation Theory. The Articulation Index was also used to compare the effectiveness of three adaptive rules. A simple multiband volume control is expected to provide much of the benefit of more sophisticated systems without the need for separate estimation of input speech and noise spectra.

An application of the articulation index to hearing aid fitting.

The application of the articulation index (AI) model to the fitting of linear amplification was evaluated for 12 subjects with sensorineural hearing loss. Comparisons were made of amplification characteristics specified by the NAL (Byrne & Dillon, 1986) and POGO (McCandless & Lyregaard, 1983) prescriptions, as well as a procedure that attempted to maximize the AI (AIMax). For all subjects, the relationship between percent-correct scores on a nonsense syllable test and AIs was monotonic for the two prescriptions, indicating that the AI was effective for comparing conditions typical of those recommended clinically. However, subjects having sloping high-frequency hearing losses demonstrated nonmonotonicity due to poor performance in the AIMax condition. For these subjects, the AIMax condition required much more gain at high than at low frequencies, circumstances that Skinner (1980) warned will cause less-than-optimal performance for individuals having sloping high-frequency hearing loss.

The relation between loudness and intensity difference limens for tones in quiet and noise backgrounds.

Recent studies of the relation between loudness and intensity difference limens (DLs) suggest that, if two tones of the same frequency are equally loud, they will have equal relative DLs [R. S. Schlauch and C.C. Wier, J. Speech Hear. Res. 30, 13-20 (1987); J.J. Zwislocki and H.N. Jordan, J. Acoust. Soc. Am. 79, 772-780 (1986)]. To test this hypothesis, loudness matches and intensity DLs for a 1000-Hz pure tone in quiet and in a 40-dB SPL spectrum level broadband noise were obtained for four subjects with normal hearing. The DLs were obtained in both gated- and continuous-pedestal conditions. Contrary to previous reports, equally loud tones do not yield equal relative DLs at several midintensities in the gated condition and at many intensities in the continuous condition. While the equal-loudness, equal-relative-DL hypothesis is not supported by the data, the relation between loudness and intensity discrimination appears to be well described by a model reported by Houtsma et al. [J. Acoust. Soc. Am. 68, 807-813 (1980)].