Acoustical Society of America and Acoustical Society of Japan Third Joint Meeting, Honolulu, Dec 2-6, 1996.



Comparison of Speech Recognition in Noise
using Laboratory and Clinical Procedures

Linda M. Thibodeau

School of Human Development, GR41
The University of Texas at Dallas
Box 830688, Richardson, Texas 75083-0688
Email: thib@utdallas.edu
Web address: http://www.utdallas.edu/~thib/

Introduction


The speech recognition threshold (SRT)in noise is a sensitive measure to quantify the effects of hearing loss (Plomp, 1978) and to evaluate aided benefit (Van Tasell and Yanz, 1984). The technique has been used primarily in the laboratory and only recently has a clinical procedure been developed for determining a threshold for sentence recognition in noise (Nilsson, Soli, Sullivan, 1994). Plomp (1978) proposed that the effects of hearing loss could be quantified by two values: "A" and "D" representing the attenuation and distortion factors, respectively. When the SRT is obtained in quiet and in several noise levels, the A and D values can be determined according to the following equation:

where Lo is the SRT for normals in quiet, Ln is the level of the noise, and Lsn is Ln minus the SRT in noise for normals.

Purpose


The purpose of this study was to evaluate an abbreviated SRT procedure from which the A and D values could be obtained to characterize the effects of hearing loss. The first experiment was conducted to determine if the standard laboratory procedure could be shortened by measuring the speech recognition threshold in only two conditions, quiet and a moderate noise level, in listeners with normal hearing. The second experiment was conducted to compare the complete and abbreviated procedures when using laboratory and clinical methods in normal and hearing-impaired listeners.

General Methods


Testing was conducted in double-walled suites with a TDH 39 earphone. Laboratory procedures were controlled by a PC utilizing a 20-kHz sampling rate. Clinical procedures were conducted using a GSI16 Diagnostic Audiometer. SRT's were obtained using the monosyllabic, high-frequency words of Van Tasell and Yanz (1984). The words were: hit, hick, hip, sit, sip, sick, thick, and ship. The competition was speech-shaped noise that was filtered from a noise generator (low-pass, 500 Hz, 12 dB per octave) in the laboratory procedure or speech noise from the audiometer in the clinical procedure.


Experiment One
Standard vs. Abbreviated Laboratory Procedures

Question:


Are there significant differences between the SRT's obtained in the laboratory when only two-conditions are used rather than five in listeners with normal hearing? Subjects: The participants were 18 students, ages 18-43, with a mean age of 24.5 years. All students passed a hearing screening at 20 dB HL for the frequencies 500 through 8000 Hz. Subjects were randomly assigned to practice (N=9) or no-practice groups (N=9).

Subjects:


The participants were 18 students, ages 18-43, with a mean age of 24.5 years. All students passed a hearing screening at 20 dB HL for the frequencies 500 through 8000 Hz. Subjects were randomly assigned to practice (N=9) or no-practice groups (N=9).

Method:


SRT's were obtained using the simple up-down procedure used by Van Tasell and Yanz (1984). Three estimates in each noise condition were averaged to obtain the final threshold. In the abbreviated procedure, thresholds were obtained in quiet and in 54 dBA of noise. In the standard procedure, thresholds were obtained in quiet and in 14, 34, 54, and 74 dBA of speech noise. The practice group received a block of trials in quiet and in the moderate noise level prior to data collection. All subjects completed the abbreviated procedure first.

Results:


The SRT's in quiet and as a function of noise are shown in Figure 1.

A repeated-measures ANOVA was performed on the SRT's obtained in quiet and moderate noise in the abbreviated and standard procedures. The only significant difference besides the expected one between conditions, was between groups for the quiet condition. However, the difference was only 3 dB which is smaller than the clinically-accepted variability in thresholds of 5 dB. To compare the abbreviated function to the standard function, the data were analyzed using Eq. 1 to derive A and D values. The entire data set was used to determine the average Lo (19.3 dB SPL) and Lsn (-8.6 dB) values which were then used to determine the A and D values as shown in Table 1. A repeated-measures ANOVA revealed a significant difference between groups only for the A value.

Table 1.
A and D Values for Abbreviated (Abb) and Standard (Sta) Procedeures
in Listeners with Normal Hearing

A Values D Values
Abb Sta Abb Sta
Practice -0.5 -1.7 -0.4 0.0
No Practice 2.5 0.6 -1.0 0.2
Overall Mean 1.1 -0.5 -0.8 0.1

Experiment Two Laboratory versus Clinical Procedures

Question:


Are there significant differences in the SRT's obtained using laboratory versus clinical procedures in two groups of listeners, those with normal hearing and those with hearing loss?

Subjects:


There were 11 participants in each subject group. Mean age of those with normal hearing was 23 years old (range 18-43) and those with hearing loss was 49 years old (range 21-75). The normal-hearing subjects all passed a hearing screening at 20 dB HL for 500 through 8000 Hz. The hearing-impaired subjects exhibited a range of sensorineural hearing losses from mild to severe. The data from three hearing-impaired subjects were excluded from the statistical analysis because of high standard deviations or minimal threshold shifts.

Method:


The procedures for the laboratory measures were the same as those used in Experiment One. The method recommended by ASHA (1988) for determining SRT was used for the clinical procedure. This involves descending to a level where two words are missed followed by two words at successive 2-dB decrements until five out of six words are missed. Threshold is calculated by subtracting the total number of correct responses from the starting level and adding a correction factor of one. The order of conditions included abbreviated and standard functions first in the clinic, then in the laboratory. For the abbreviated function, only one threshold estimate was obtained. For the full function, three threshold estimates were averaged to obtain the final threshold. The same noise levels from Experiment One were used.

Results:

Standard Function:


Figure 2 illustrates the full speech recognition functions for the normal and hearing-impaired listeners in the clinical and laboratory conditions.

The SRT's obtained in the clinic and the laboratory were significantly correlated across all subjects and all conditions (r=.97) as shown in Figure 3. The correlations for the SRTıs of the impaired subjects in each noise condition ranged from .86 to .95. Table 2 shows the A and D values.

Table 2.
A and D Values obtained in Laboratory and Clinical Settings
for Normal and Hearing Impaired Listeners.
Standard Deviations are in Parenthesis.

A Values D Values
Lab
Normals 5.2 (3.1) -0.5 (2.0)
Impaired 19.8 (8.5) 3.1 (6.0)
Clinic
Normals 2.4 (3.9) 2.1 (.9)
Impaired 17.2 (8.3) 7.8 (6.5)

A repeated-measures ANOVA for the standard functions revealed significant differences between the clinical and laboratory data for both D and the A values. However, the A and D values obtained in the lab and the clinic were significantly correlated (A:r=.96, D:r= .79). Abbreviated Function: The SRT's in the quiet and moderate noise level are shown in Figure 4. The SRT's were significantly correlated at both the quiet (r=.97) and the moderate noise level (r=.53). The A and D values determined from the abbreviated functions were also significantly correlated (A:r= .79, D:r= .80).

Figure 3. Correlations between SRT's obtained in the lab and the clinic in Normal and Hearing-impaired listeners.

Results:

Abbreviated Function:
The SRT's in the quiet and moderate noise level are shown in Figure 4. The SRT's were significantly correlated at both the quiet (r=.97) and the moderate noise level (r=.53). The A and D values determined from the abbreviated functions were also significantly correlated (A:r= .79, D:r= .80).

Figure 4. Abbreviated SRT Functions for Normals and Hearing-Impaired Listeners Obtained in the Laboratory and Clinical Settings.

Summary


SRT's obtained in quiet and in noise conditions can provide a useful representation of one's auditory ability in terms that reflect attenuation (A) and distortion (D) (Plomp, 1978). An abbreviated SRT procedure yielded equivalent data within the range of acceptable clinical variation. The data obtained using the abbreviated procedure were significantly correlated with that obtained in the laboratory setting in listeners with normal and impaired hearing. Further evaluation of the abbreviated clinical procedure is warranted to determine it's sensitivity as a clinical measure of speech recognition.

References

American Speech, Language, and Hearing Association. (1988). Guidelines for
determining threshold level for speech. ASHA, 30, 85-90.
Nilsson, M., Soli, S., and Sullivan, J. (1994). Development of the hearing
in noise test for the measurement of speech reception thresholds in quiet and in noise. Journal of the Acoustical Society of America, 95, 1085-1096.
Plomp, R. (1978). Auditory Handicap of hearing impairment and the
limited benefit of hearing aids.
Journal of the Acoustical Society of
America, 63,533-549.
Van Tasell, D. and Yanz, J. (1985). Speech recognition threshold in noise:
Effects of hearing loss, frequency response, and speech materials.

Acknowledgments


This research was supported by NIDCD.


Dr. Thibodeau's Home Page | Speech Perception Lab Page