Frequency Selectivity and Frequency Discrimination

2 .2 Signal Processing Strategies for the Profoundly Hearing-Impaired

with Residual Low Frequency Hearing

2.2.1 Frequency Recoding Hearing Aids * Frequency Compression

* Energy Shifting Aids * Spectrum Shifting Aids

CHAPTER 2 SIGNAL PROCESSING FOR HEARING IMPAIRED PEOPLE WITH SEVERE HEARING LOSS AT HIGH FREQUENCIES

CHAPTER 2

SIGNAL PROCESSING FOR HEARING-IMPAIRED PEOPLE

WITH SEVERE HEARING LOSS AT HIGH FREQUENCIES

In order to develop effective rehabilitative methods for hearing-impaired people, the relationships between hearing and speech perception must be studied and applied. If the impaired perception of phonetic features can be accounted for in terms of reduced performance with specific acoustic cues, then this should enable individual deficits in speech discrimination and recognition to be described in more detail and, therefore, this knowledge could help to improve prostheses for this group. Thus this chapter begins with the examination of hearing impairment and the consequent disability in speech perception; then various methods of speech processing based on the details of, and designed to reduce, these perceptual deficits will be described.

2.1 Aspects of Sensori-Neural Hearing Impairment

This section deals with auditory perception in those with sensori-neural hearing loss, but omits discussion of conductive hearing loss, since pure conductive losses are often successfully treated with amplification or in some cases surgery and do not necessitate further aids to speech understanding.

2.1.1 Sensitivity for Detection

Evaluation of detection thresholds for pure tones ( pure-tone audiogram) is the most commonly used test for hearing impairment. Thresholds for pure tones are expressed in dB HL (decibels hearing level) and are calibrated such that the average threshold for a normally hearing person is at 0 dB for frequencies from 0.125 -1 0 kHz. The audiograms of the hearing-impaired listeners used for perceptual tests later in this

Study are shown in Appendix C. It can be seen that sensori-neural hearing losses are likely to be worse in the higher frequencies. Since most consonants have their contrastive information (e.g. formant transitions) in the high frequency region, people with a sensori-neural hearing loss will have greater difficulties in perceiving these major carriers of lexical information in the English language (Fry, 1976).

The sensitivity loss can be largely corrected by conventional amplification hearing aids. But, for a number of reasons, speech discrimination scores depend only partially upon sensitivity levels, and one has to look at other impaired perceptual processes in order to understand the complex phenomena occurring in those with sensori-neural hearing losses.

2.1.2 Intensity Coding

There is a dynamic range of hearing which lies between the minimum-audibility threshold and the loudness discomfort level (LDL) (BaUantyne & Martin, 1984). LDL is at approximately 100 dB SPL across the frequency range in normal ears. In cochlear hearing loss, detection thresholds are elevated but LDL may remain at normal or near­ normal levels. This results in a much reduced dynamic range, and as a consequence, a

diagnostic sign of cochlear hearing loss is the phenomenon of recruitment, or abnormal

loudness g ro w th. One hypothesis for the explanation of this phenomenon is related to

the marked reduction of the number of functioning receptor cells which is common in cochlear pathology (Evans, 1975).

2.1.3 Frequency Selectivity and Frequency Discrimination

Frequency selectivity denotes the ability of the auditory system to resolve or

separate out the individual spectral components of a complex signal. The loss of frequency selectivity relates to the broadening of auditory filters. Sensori-neural hearing impaired patients usually have broader auditory filter bandwidth than normal hearing and

CHAPTER 2 SIGNAL PROCESSING FOR HEARING IMPAIRED PEOPLE WITH SEVERE HEARING LOSS AT HIGH FREQUENCIES

conductive hearing impaired subjects. A particular example of how this damage may be induced is by the use of ototoxic drugs such as Kanamycin. Appendix C indicates that most of the hearing-impaired Chinese subjects used in this study were injected with these drugs at an early age.

One effect of the consequences of impaired frequency selectivity is to impair the pitch perception of complex tones, and hence, the perception of intonation in speech signals. Rosen and Fourcin (1986) give a detailed discussion of how the widening of the auditory bandwidth affects the temporal representation of fundamental frequency which is thought to be important for pitch perception (Moore & Glasberg, 1986).

The theory of pitch perception for complex tones that Moore and Glasberg (1986) detail is directly applicable to the perception of the pitch of voiced speech sounds. In their model, the pitch of complex tones is based primarily on a temporal process preceded by an initial frequency analysis, i.e., each temporal processing chaimel operates on the information contained in a selected frequency range defined by an associated auditory filter.

For normal listeners, at low frequencies (where the filters have the narrowest bandwidths) each Hlter responds primarily to a single harmonic since the harmonic spacing is wide compared to the filter bandwidth. The temporal information in the waveforms is then extremely simple and well defined by the peaks ( or valleys) of the sinusoids. The high-frequency channels, however, are excited by a number of harmonics at the same time and more complex waveforms can often be seen. It appears that the lower resolvable harmonics dominate the perceived pitch (Plomp, 1967; Ritsma,

1967).

For some hearing impaired listeners, most of the filter bandwidths are greater than the harmonic spacing. As a result the waveforms for most charmels are, temporally speaking, complex. We should therefore expect a reduced ability to discriminate

fundamental frequency changes in complex tones consequent upon reduced frequency selectivity (Rosen and Fourcin, 1986).

Speech perception is based on acoustic features that occur simultaneously, such as the spectral formants which are present at different frequencies and which are used as cues for phonetic discrimination. The reduction of frequency selectivity caused by the widening of the auditory filter bandwidth will also limit the ability of the ear to resolve the individual formants of the speech signal, and especially the ability to follow speech in noisy situations.

F requency discrim ination is the ability of the auditory system to perceive

frequency changes. The smallest detectable change in frequency is defined as the

frequency difference limen (DLF), and used as a measure of frequency discrimination.

The DLFs of profoundly hearing impaired listeners are generally much larger than those of normal listeners (Grant, 1987b). In his experiment, the DLFs for the hearing- impaired subjects were approximately 36 times larger than those for normal-hearing subjects under the condition where the amplitudes of the frequency-modulated test tones were randomly modulated.

Measures of frequency discrimination are often thought to be closely related to the ability to perceive pitch in speech ( Hoekstra & Ritsma, 1977), since the deterioration in frequency discrimination may cause difficulty in following changes in Fx.