measures based upon pure tone air-conduction thresholds (e.g., Davis, ... acknowled e the contribution of David Wong, Wayne Wong, Joan Coren, Geof Donelly, Lynda. Bereer. an3 Dereck Acha who assisted in the collection of these data.
PercephralandMotor Skills, 1994, 79, 1003-1008. O Perceptual and Motor Skills 1994
PREDICTING SPEECH RECOGNITION THRESHOLDS FROM PURE TONE HEARING THRESHOLDS ' STANLEY COREN AND A. RALPH HAKSTIAN University of British Columbia, Kzncouuer Summary.-Hearing sensitivity is most commonly stiU reported in terms of pure tone thresholds. Unfortunately, simple procedures for predicting Speech Recognition Thresholds from Pure Tone Thresholds are not currently available. To remedy this problem, pure tone thresholds were collected from 802 individuals over the range of 250 to 8000 Hz. Five subsets of pure tone thresholds which are commonly used to report hearing status were then considered. An average correlation of 0.878 was found between the various pure tone indexes and the speech recognition threshold. Using regressions between pure tone and the speech measure, a table was constructed that allows conversion of the various pure tone indexes to a predicted speech recognition threshold and involves only a very simple computation.
The most commonly used means of hearing assessment stU involves measures based upon pure tone air-conduction thresholds (e.g., Davis, 1983). Since the practical goal of testing is usually to predict "hearing handicaps" in the form of speech comprehension, various subsets of pure tone thresholds (usually centering upon the middle of the range of frequencies most needed for speech recognition) are commonly used and reported in the literature Some particular combinations of tones have even been officially adopted by various governments and private organizations for the purpose of reporting hearing ability. One example of such an "official" index is the American Medical Association (AMA) formula, which is the average of the pure tone thresholds for the frequencies 0.5, 1, 2, and 4 kHz (AMA, 1947). The AMA index has been adopted by seven states in the U.S. and also by Australia and Italy. Another widely used system is one that involves use of the frequencies 0.5, 1, and 2 kHz. This tonal combination was advocated for many years by the American Academy of Ophthalmology and Otolaryngology (AAOO) and remains the standard measure for hearing handicap in 14 states in the U.S. and also Canada, Denmark, Iran, Israel, Japan, and Brazil (AAOO, 1969). Traditionally, audiological researchers in the United Kingdom have used a formula with the frequencies of 0.5, 1, and 3 kHz. Recent recommendations from the British Association of Otolaryngologists and the British Society of Audiology include the use of a formula based upon 1, 2, and 4 kHz (the 'This research was supported in part by rants from the British Columbia Health Care Research Foundation and the Natural Sciences a n c f ~ n ~ i n e e r i nResearch g Council of Canada. The authors acknowled e the contribution of David Wong, Wayne Wong, Joan Coren, Geof Donelly, Lynda Bereer. a n 3 Dereck Acha who assisted in the collection of these data. For reorincs or other correspoid;nce, contact Dr. Stanley Coren, Department of Psychology, ~ n i v e r s hof British ~ o l u m bia, 2136 West Mall,Vancouver, B.C., Canada V6T 124.
1004
S. COREN & A. R. HAKSTIAN
BAOL/BSA formula), which is now achieving broad acceptance in the UK (BAOLIBSA, 1983). A system based upon the frequencies 0.25, 0.5, 1, 2, and 4 kHz which has been used as the measure in a number of industrial settings, court cases, and VA hospitals is one we refer to as the "Industrial Index" (Noble, 1978; Suter & von Gierke, 1975). Coren (1989), on the basis of empirical evidence consisting of the intercorrelations among the pure tone thresholds for six of the most commonly used audiometric test frequencies (0.25, 0.5, 1, 2, 4, and 8 kHz), recommended that a Full Range Average of these frequencies be used. Although the Full Range Average has not usually been used to predict speech recognition, it does appear to be a good summary measure of the information in the audiogram since it has a mean correlation of 0.89 with the tonal threshold for any of the individual audiometric frequencies (Coren, 1989). It is not our purpose to resolve the issue of which of these indexes (or any of the myriad of other combinations of test frequencies not specifically mentioned here) is the best measure for predicting hearing handicap for speech. Instead we address a more pragmatic problem, namely, "Is it possible simply to predict the speech recognition threshold from such pure tone measures?" This is a problem which is encountered by researchers who must estimate the intelligib~lityof speech to various groups or to individuals who have undergone particular experimental treatments when the only data available are tonal thresholds. The problem is complicated, first by the fact that each auditory researcher will tend to select one of the various popular combinations of test tones to report, hence the available subsets of pure tone measures will tend to vary across published reports and from different laboratories. Secondly, one cannot simply use the average pure tone thresholds as an estimate of speech recognition, since this value will be 12 to 20 dB lower than the value that might be obtained if the individual were directly tested using spoken stimuli, although the precise conversion factors from the tonal to the speech recognition measures for each index are not known. Coren and Hakstian (1990a) recently addressed the problem of conversion from one measure of threshold sensitivity to another given the fact that each may be based upon a separate subset of frequencies. They showed that it is possible to develop simple conversion formulae which allow measures presented in terms of any one of the pure tone threshold indexes to be transformed directly into any other. Using similar techniques, it should be possible to develop empirical formulae which would allow conversion from any set of pure tone hearing threshold measures directly to speech recognition thresholds. This would then allow direct estimation of speech recognition even if only pure tone measures are available, and it would also allow comparison across studies which use different subsets of pure tones to measure hearing sensitivity. To test the feasibility of this, we conducted the following experiment.
PREDICTING SPEECH RECOGNITION THRESHOLDS
1005
METHOD Subjects The sample consisted of 802 subjects (479 women and 323 men) who responded to advertisements in a university setting and in the general community, and whose numbers were augmented by direct solicitation in several groups of seniors. Subjects ranged in age from 17 to 92 years, with a mean age of 36.2 yr. The gouped age distribution was as follows: 16 to 35 years, 426; 36 to 55, 159; 56 and older, 217. Information on occupation was not available for this sample.
Procedure
AU subjects were tested individually in a sound-deadened room. A MAICO (MA-24) audiometer was used to obtain pure tone air-conduction hearing thresholds for six test frequencies, 250, 500, 1000, 2000, 4000, and 8000 Hz. Each ear was tested separately using an initial descending sequence of tones, followed by three ascending sequences with 5-dB steps. The median of the three ascending measures served as an estimate of threshold sensitivity for that ear. Test stimuli for speech recognition were spondees (which are two-syllable words with equal emphasis on both syllables, such as "baseball" or "blackboard"). These were spoken by a male voice and were from the W-l tape commonly used for speech recognition threshold testing in many clinical settings, and the method for assessing speech recognition followed the procedures often used clinically (Newby & Popelka, 1985). Starting at a level 30 dB above the pure tone threshold for 1000 Hz, the test level was decreased in 5-dB steps for each correct response. After two consecutive misses, the test level was raised 10 dB. Five spondees were given at each level, decreasing in 5-dB steps until a level was reached where all stimuli were missed. The 50 percent intelligibility level then served as the speech recognition threshold. RESULTSAND DISCUSSION Since the treatment of the thresholds from the two ears of the same individual as independent may lead to statistical confounding (see Coren & Hakstian, 1990b), the data analyses were arbitrarily limited- to results from the left ear of each subject. A preliminary consideration of the data indicated a broad range of pure tone threshold sensitivity in this heterogeneous sample. When hearing thresholds were averaged across the six test frequencies, values ranged from a minimum value of - 1.69 to a maximum of 110.0 dB SPL. Since various reports in the literature use the different combinations of pure tones described above to report their results, we decided to compute first each of the five indexes (AAOO, BAOLIBSA, AMA, Industrial, and
1006
S. COREN
&
A. R. HAKSTIAN
Full Range) for each subject. The first question asked was how closely these indexes were correlated with the Speech Recognition Threshold. To this end, Pearson product-moment correlations were computed between each index and the speech recognition threshold. These correlations are presented in Table 1, along with the means and standard deviations for each measure. TABLE 1 PEARSON CORRELATIONS BETWEENHEARING HANDICAP INDEXES (BASEDUPON PURETONE AVERAGES) AND SPEECH RECOGNITION THRESHOLDS (BASEDUPONW-1 SPONDEES) Pure Tone Hearing Handicap Index AAOO Industrial AM A
F d Range BAOL/BSA
Correlation With Speech Recognition Threshold
Mean for
Sample
Standard Deviation for Sample
,893 ,889 .881 .870 ,855
14.74 15.71 14.78 16.86 13.05 33.20
12.14 13.06 14.32 15.46 16.41 12.98
Speech Recognition Threshold Note.-N = 802; all correlations are significant with p
