Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2009 Mar 19.
Published in final edited form as: Proc Meet Acoust. 2008 Sep 3;2(1):1–9. doi: 10.1121/1.2988004

Preliminary evaluation of the speech perception assessment and training system (SPATS) with hearing-aid and cochlear-implant users

James D Miller a, Charles S Watson a, Doris J Kistler b, Frederic L Wightman b, Jill E Preminger c
PMCID: PMC2658626  NIHMSID: NIHMS73423  PMID: 19305636

I. INTRODUCTION

Numerous recent studies have demonstrated improved speech perception by hearing-impaired listeners after training (Burk et al., 2006; Burk and Humes, 2007; Burk and Humes, 2008; Fu, 2007; Miller et al., 2004; Miller et al., 2005; Stecker et al., 2006; Sweetow and Henderson-Sabes, 2004; Sweetow and Palmer, 2006; Sweetow and Sabes, 2006). This study was conducted as a preliminary evaluation of the usefulness of the Speech Perception Assessment and Training System (SPATS, patent pending) with hearing-aid and cochlear-implant users. The methods and results are described. Additional details of the SPATS program are given by Miller et al., 2007.

II. METHODS AND PROCEDURES

A. Participants

Twelve hearing-aid users and 16 cochlear implant users served as paid participants in this study. The hearing-aid users were recruited in Bloomington, IN and they are described in Table 1. The cochlear-implant users were recruited at the Heuser Hearing Institute in Louisville, KY and are described in Table 2. These samples include a wide range of ages, types of hearing-loss, and histories.

TABLE 1.

Hearing-aid users.

Number Group Age SRT (best ear) Speech (best ear) Audiogram (best ear) Hearing Loss
520 Trained 26 60 dB 96% at 90 dB Flat Sensorineural
517 Trained 71 47 not avail. Sloping Sensorineural
519 Trained 76 30 80% at 70 dB High Tone Sensorineural
511 Trained 79 30 80% at 70 dB High Tone Sensorineural
503 Trained 87 0 80% at 80 dB Sloping Sensorineural
504 Trained 87 20 84% at 69 dB Steep Slope Sensorineural
516 Trained 88 45 68% at 85 dB Sloping Sensorineural
513 Trained 90 85 72% at 100 dB Flat at 65 dB Mixed at Lows
514 Control 69 25 94% at 65 dB Gentle Slope Sensorineural
506 Control 70 15 96% at 55 dB Mild High Tone Sensorineural
502 Control 86 30 76% at 70 dB Flat/Slope Sensorineural
505 Control 86 55 88% at 95 dB Gentle Slope Sensorineural
Open in a new tab

TABLE 2.

Cochlear-implant users.

Number Group Age Onset Age Years w/CI Processor
620 Trained 35 26 1 Freedom
619 Trained 39 37 2 Freedom
504 Trained 42 3.5 4 Freedom
616 Trained 57 1.5 1 Freedom
610 Trained 69 64 5 Freedom
611 Trained 70 55 1 Freedom
613 Trained 70 41 2 Freedom
624 Trained 81 45 4 Sprint
617 Control 42 Early Childhood 4 3G
622 Control 42 Early Childhood 5 Freedom
623 Control 43 3 2 Freedom
625 Control 43 3 mos. 3 Freedom
628 Control 49 20 1 Freedom
627 Control 63 6R, 40L 3 Freedom
618 Control 64 44 6 Sprint
626 Control 79 59 19 3G
Open in a new tab

B. Pre- and Post-Tests

The participants were pre-tested on a variety of non-SPATS tests and then on a variety of SPATS tests. After pre-testing the controls were given a several week hiatus, while the trained participants were scheduled for 6 weeks of training (two two-hour sessions each week). Finally, all participants were post-tested on SPATS tests and then on non-SPATS tests. Table 3 lists the non-SPATS tests used for the HA-users and the CI-users. These were administered both before and after training for the trained participants and before and after a several week hiatus for the control participants. All speech materials were presented at 65 dB SPL.

TABLE 3.

The non-SPATS tests given to HA- & CI-users are listed.

HA Test W22 W22 HINT HINT CST-AV* CST-A*
Users SNR (Quiet) + 8 dB (Quiet) + 8 dB −4 dB − 4 dB
CI Test CNC HINT HINT
Users SNR (Quiet) (Quiet) +10 dB
Open in a new tab
*

AV = Auditory-Visual, A = Auditory-Alone

The SPATS tests are listed in Table 4. The first row gives the combinations of constituent types and levels. For each combination, one test was run in the quiet and the other was adaptive in noise. Adaptive testing and training in noise (12-talker babble) was done using Kaernbach’s weighted up-down procedure (Kaernbach, 1991) to find the SNR for a target percent correct (TPC). The target percent correct (TPC) was chosen as 10% below the listener’s score in quiet. (This rule has been abandoned in current versions of SPATS as it is clinically inappropriate to ask clients to adapt to target scores below 50% correct.) Sentence testing was done with the Sentence Module (patent pending). One test was done with no sound to test the individuals’ abilities to guess the words in sentences. A second sentence test was done auditorily but with the SNR adapted in accordance with the subjects’ performance. These procedures are described in more detail in the companion paper Miller et al., 2007. Nuclei Level 1 consists of the 7 most important vowels of English. Nuclei Level 3 has the 21 most important vowel and vowel-like sounds of English. Onsets Level 1 consists of the 11 most important beginnings of English words, while Onsets Level 3 consists of the 34 most important beginnings of English words. (“Importance” was scaled in terms of frequencies of occurrence, see companion paper for details.) In a test each item in the set is presented 4 times, with talker and vowel chosen randomly. For onsets each item in a set was recorded by 8 talkers as combined with each of four vowels (/i/, /ɑ/, /u/, /ɚ/). The vowels were also recorded by 8 talkers but in only one context, hVd.

TABLE 4.

List of SPATS Tests given to HA- and CI- users.

Nuclei Level 1 Nuclei Level 3 Onsets Level 1 Onsets Level 3 Sentences
Quiet Quiet Quiet Quiet No Sound
Adapt. to TPC* Adapt. to TPC Adapt. to TPC Adapt. to TPC Adaptive
Open in a new tab
*

TPC is target percent correct

C. Training

Training was organized as a “training rotation.” The goal was to complete 12 rotations, where each rotation included training on the identification of syllable onsets and nuclei in quiet and in noise and training in identifying short sentences at signal-to-babble ratios that were adapted to each participant’s performance. A training rotation is summarized in Table 5.

TABLE 5.

One rotation of training is shown.

Sequence Constituents Level SNR # Items #Trials Est. Time Minutes Cum. Time Minutes
1 Nuclei 1 Quiet (+40 SNR) 7 28 3 3
2 Nuclei 1 SNR-Adapt. To TPC 7 28 3 6
3 Sentences n/a SNR-Adaptive 3 sets of 3 sent. 9 15
4 Nuclei 3 Quiet (+40 SNR) 21 84 10 25
5 Nuclei 3 SNR-Adapt. To TPC 21 84 10 35
6 Sentences n/a SNR-Adaptive 3 sets of 3 sent. 9 44
7 Onsets 1 Quiet (+40 SNR) 11 44 5 49
8 Onsets 1 SNR-Adapt. To TPC 11 44 5 54
9 Sentences n/a SNR-Adaptive 3 sets of 3 sent. 9 63
10 Onsets 3 Quiet (+40 SNR) 34 136 20 83
11 Onsets 3 SNR-Adapt. To TPC 34 136 20 103 (1 hr 43 m)
Open in a new tab

There are several characteristics of the training rotation worthy of comment. For syllable constituents, training runs used the adaptive item selection (AIS) algorithm (patent pending) described in the companion paper. AIS is designed to focus training on learnable items of intermediate difficulty. Levels 1 and 3 were selected for constituent training so that the participants would alternate between easier and more difficult tasks as means of maintaining motivation. The sentence task was intermixed with the constituent tasks. Sentence scores calculated after each group of three sentences were used to adjust the SNR for the next set of three sentences. With breaks included most participants were able to complete one rotation in a two-hour session. Three of the trained HA users completed six rotations or about 12 hours of training, as other obligations prevented them from completing all 12 rotations. All other trained HA and CI participants completed the 12 planned rotations for approximately 24 hours of training.

III. RESULTS

A. SPATS Tests

1. Hearing-Aid Users--Constituents in Quiet

The gains made at Levels 1 & 3 for onsets and nuclei varied from one individual to the next. Therefore, scores are presented for individual participants averaged across level (1 and 3) and constituent type (nuclei and onsets). The scores for individual HA users are presented in Figure 1 below; the upper panels show the results for testing in quiet. For each participant the average score on the pretests is shown by the red bar, the average score on the posttests by the black bar, and the gain by the green bar. The average gain for trained participants was about 8%, while that for controls was 0%. Note the individual differences.

FIGURE 1.

FIGURE 1

Open in a new tab

Results for HA-users on SPATS pre- and post-tests. Left panels are for trained participants and the right for control participants. The results are averages of the data for 4 conditions: Nuclei L1 & L3 and Onsets L1 & L3.

2. Hearing-Aid Users--SNR Adaption

The bottom panels of Figure 1 show the participants’ performance in noise in terms of % of Norm (%Norm). For every target percent correct and constituent type and level the signal-to-noise ratio obtained by normal-hearing listeners (SNRN) was determined. A client’s signal-to-noise ratio (SNRC) was measured. The %Norm for an individual was calculated by Eq. 1.

%Norm=((SNRC−40)/(SNRN−40))∗100, (Eq. 1)

where the constant 40 represents effective quiet. If a client requires an SNR of 40 to achieve their target percent correct, they are unable to hear down into noise and their %Norm is zero. If the Client’s SNRC equals the SNRN, their %Norm is 100. While there is individual variation, the average %Norm on pretest was 60%, while after training this number increased to 75% for a gain of 15%. The controls averaged about 73% on the pretest and fell to about 71% on the posttest for a rounded loss of 1%. These changes in %Norm for the trained group correspond to an average reduction in the SNR required to reach the target percent correct of 7.3 dB.

3. Cochlear-Implant Users—Constituents in Quiet & SNR Adaption

Figure 2 shows the results of constituent tests for the cochlear-implant users. In the upper-left panel it can be seen that two of the trained CI users showed substantial gains in percent correct in the Quiet, while the others did not. The mean improvement for the trained implant users was 6%, while for controls it was 0%. The improvements in noise were more consistent, with only one participant1 showing no improvement. Indeed, the average improvement for the trained group in %Norm was almost 14%, while for controls it was 2%. These changes in %Norm for the trained group correspond to an average reduction in the SNR required to reach the target percent correct of 8 dB. Note the changes in %Norm for Participant 610. For those conditions where 610’s target percent corrects (TPCs) were reasonable percentages, e.g. near 50%, this subject showed in gains in %Norm of about 20%. When the TPCs were very low, e.g. about 15%, this subject showed gains in %Norm of about −25%. In retrospect, it was unreasonable to ask the subject to track at such very low percentages.

FIGURE 2.

FIGURE 2

Open in a new tab

Results for CI-users on SPATS pre- and post-tests. Left panels are for trained participants and the right for control participants. The results are averages of the results for 4 conditions: nuclei l1 & l2 and onsets l1 & l2.

4. Average Results for HA- and CI-users

The average results on SPATS constituent tests for HA- and CI-users are shown in Table 6. It is very clear that there were no gains for the controls and that trained participants improved by an average of 7% in quiet and 15% in terms of %Norm in noise. If one averages these scores, the overall average gain was 11% for the trained group and 0% for the controls, the former being a highly significant improvement. The t-values for the paired pre- and post-test measures were 10.0894 for the trained group (df =15) and 0.3233 for the control group (df =11), while the corresponding critical t-values (α =0.05 for two-tailed tests) were 2.1314 and 2.2010, respectively. The correlations between pre- and post-tests were 0.90 for the trained group and 0.96 for the control groups, indicating that the scores are reliable and that training does not substantially change performance rankings within a group.

Table 6.

SPATS constituent measures combined for HA- and CI-users.

Group Quiet Percent Correct SNR-Adaption % Norm Mean
Pre Post Gain Pre Post Gain Pre Post Gain
Trained 55 62 7 57 71 15 56 67 11
Control 59 59 0 66 67 0 63 63 0
Open in a new tab

5. Performance on the SPATS sentence tasks

In the course of training, the average participant, HA- or CI-user, listened to and successfully identified 367 sentences which combined to total 2040 words. About 24% of the total training time was spent working on the sentence task. Almost all HA-users showed improvements in understanding the sentence material at signal-to-nose ratios of −5 to −15 dB. At this time, similar analyses have not been completed for CI users. No sentence data are shown for either group as we are in the process of revising the sentence task and developing appropriate measures that will allow comparisons to the performances of normally hearing peers. Nonetheless, it is believed that the sentence task is a necessary and key component of SPATS as it trains almost all of the relevant bottom-up and linguistic top-down skills involved in speech perception. All but one or two of the participants reported enjoying the sentence task and found it relevant to everyday hearing of speech. There were fairly large individual differences in their abilities to use context in this task.

B. Non-SPATS Tests

1. Hearing-Aid Users

The results for the hearing-aid users on non-SPATS tests are shown in Table 7. The last row of the table shows the mean difference between the trained and the control groups. Averaged over all measures the trained subjects improve by 8 percentage points more than the controls. However, for most of the participants the HINT scores were too high to allow significant improvement. Also, the CST-A-Only task was too difficult at the selected SNR. When the HINT and CST-A-Only scores are ignored, the trained subjects improved from pre- to post-tests by about 10 percentage points more than the controls. Overall, the improvements on non-SPATS measures were about equal to the improvements on SPATS measures.

Table 7.

Non-SPATS measures for hearing-aid users. All scores are in percent correct.

Tests W22 W22 HINT HINT CST-AV CST-A-Only Mean
Quiet +8 dB SNR Quiet +8 dB SNR −4 dB SNR −4 dB SNR Non-SPATS
Group Particpants Pre Post Gain Pre Post Gain Pre Post Gain Pre Post Gain Pre Post Gain Pre Post Gain Pre Post Gain
Trained 503 50 64 14 6 44 38 87 100 13 79 89 10 32 38 6 2 4 2 43 57 14
Trained 516 66 66 0 8 32 24 93 99 6 99 98 −1 13 20 7 4 3 −1 47 53 6
Trained 517 66 82 16 20 74 54 98 97 −1 92 92 0 6 5 −1 0 0 0 47 58 11
Trained 519 60 70 10 62 48 −14 85 91 6 75 80 5 12 22 10 2 1 −1 49 52 3
Trained 520 72 86 14 46 72 26 94 98 4 100 95 −5 42 55 13 8 14 6 60 70 10
Trained 504 70 84 14 26 62 36 100 99 −1 99 100 1 38 48 10 15 21 6 58 69 11
Trained 511 60 58 −2 28 64 36 99 100 1 97 97 0 61 68 7 23 22 −1 61 68 7
Trained 513 24 56 32 1 24 23 89 100 11 91 88 −3 13 24 11 4 5 1 37 50 13
Trained Mean 59 71 12 25 53 28 93 98 5 92 92 1 27 35 8 7 9 2 50 60 9
Control 502 56 70 14 14 34 20 99 100 1 98 99 1 55 50 −5 5 4 −1 55 60 5
Control 505 60 58 −2 8 36 28 96 97 1 89 92 3 27 33 6 4 3 −1 47 53 6
Control 506 94 98 4 78 80 2 100 99 −1 99 99 0 83 60 −23 36 9 −27 82 74 −8
Control 514 88 96 8 76 88 12 100 100 0 100 100 0 39 39 0 17 6 −11 70 72 2
Control Mean 75 81 6 44 60 16 99 99 0 97 98 1 51 46 −6 16 6 −10 63 65 1
T-C Mean −16 −10 6 −19 −7 12 −6 −1 5 −5 −5 0 −24 −11 13 −8 3 12 −13 −5 8
Open in a new tab

2. Cochlear-Implant Users

The results for cochlear-implant users on non-SPATS tests are shown in Table 8. The last row of the table shows the mean differences between the trained and control groups. Training seemed to produce the greatest gains for HINT sentences in quiet. Overall the trained group improved 13% relative to the control group.

Table 8.

Non-SPATS measures for cochlear-implant users. All scores are percent correct.

Tests CNC HINT HINT Mean
Quiet Quiet +10 dB SNR Non-SPATS
Group Particpants Pre Post Gain Pre Post Gain Pre Post Gain Pre Post Gain
Trained 504 84 88 4 98 100 2 57 78 21 80 89 9
Trained 610 44 50 6 60 85 25 41 59 18 48 65 17
Trained 611 24 26 2 71 62 −9 30 42 12 42 43 1
Trained 613 64 66 2 76 73 −3 25 27 2 55 55 0
Trained 616 20 34 14 21 63 42 19 19 0 20 38 18
Trained 619 48 62 14 78 98 20 60 69 9 62 76 14
Trained 620 44 50 6 62 83 21 35 61 26 47 65 18
Trained 624 48 44 −4 81 87 6 57 48 −9 62 60 −2
Trained Mean 47 53 6 68 81 13 41 50 10 52 61 9
Control 617 62 48 −14 92 77 −15 63 32 −31 72 52 −20
Control 618 28 30 2 55 53 −2 15 23 8 33 35 2
Control 622 40 46 6 66 54 −12 35 31 −4 47 44 −3
Control 623 28 32 4 25 28 3 20 7 −13 24 22 −2
Control 625 70 66 −4 92 93 1 69 79 10 77 80 3
Control 626 60 48 −12 84 67 −17 47 62 15 64 59 −5
Control 627 70 66 −4 92 93 1 69 79 10 77 79 2
Control 628 60 48 −12 85 67 −18 47 62 15 64 59 −5
Control Mean 52 48 −4 74 67 −7 46 47 1 57 54 −3
T-C Mean −5 5 10 −6 15 20 −5 4 9 −5 8 13
Open in a new tab

3. Non-SPATS Summary

If the results for all tests and both HA- and CI-users are averaged, it is found that the trained participants had a pretest average of 51% and a posttest average of 60% for a gain of 9%. The control participants had a pretest average of 52% and a posttest average of 51% for a loss of 1%. Overall the trained participants improved 10% relative to the controls. The t-values for the paired pre- and post-test measures were 5.9433 for the trained group (df =15) and 0.8452 for the control group (df =11), while the corresponding critical t-values (α =0.05 for two-tailed tests) were 2.1314 and 2.2010, respectively. The correlations between pre- and post-tests were 0.88 for the trained group and 0.92 for the control groups, indicating that tests are reliable and that training does not substantially change score rankings within a group.

IV. SUMMARY AND CONCLUSIONS

Experienced hearing-aid and cochlear-implant users demonstrate improvements in speech perception after 12 to 24 hours of training on the SPATS system. This improvement is demonstrated by SPATS measures and by clinical audiologic tests in quiet and in noise. Participants ranged in age from 26 to 90 years. All participants, including the CI users, were capable of maneuvering through the SPATS program in a fairly independent manner. A wide range of hearing losses and hearing loss histories were included. The adaptive nature of the program allowed all participants to achieve enough correct identifications during training in order to keep motivation levels high. Overall, the trained participants showed gains of 11% over the controls. The controls showed no improvement. HA users commented that the program helped them to understand their hearing losses. They said that they became aware of their specific problems in the identification of speech sounds, and that they learned to attend to the details of sounds and to talker differences.

Acknowledgments

James D. Miller and Charles S. Watson are stockholders in Communications Disorders Technology, Inc. and could profit from the commercialization of SPATS software. This work was supported by NIH/NIDCD by SBIR Grant R44DC006338.

Footnotes

1

The results for participant #610 illustrate the effects of asking a client to track to very low TPCs. As shown in the bottom left-hand panel of Figure 2, the average gain for 610 was negative and small. For two of the four conditions (Onsets and Nuclei at Level 1) his TPCs were near 50% and for these conditions he showed gains in %Norm with training of about 20%. For the other two conditions (Onsets and Nuclei at Level 3), his TPCs were near 15% and his gains were about −25%. Generally, it has been observed that tracking to very low TPCs was both confusing and discouraging to the participant.

References

  1. Burk M, Humes LE. Effects of training on speech recognition performance in noise using lexically hard words. J Sp Lang Hear Res. 2007;50:25–40. doi: 10.1044/1092-4388(2007/003). [DOI] [PubMed] [Google Scholar]
  2. Burk M, Humes LE. Effects of long-term training on aided speech-recognition performance in noise in older adults. J Sp Lang Hear Res. 2008;51:759–771. doi: 10.1044/1092-4388(2008/054). [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Burk M, Humes LE, Amos NE, Strauser LE. Effect of training on word recognition performance in noise for young-normal hearing and older hearing-impaired listeners. Ear & Hear. 2006;27:263–278. doi: 10.1097/01.aud.0000215980.21158.a2. [DOI] [PubMed] [Google Scholar]
  4. Fu QJ, Galvin JJ. Perceptual learning and auditory training in cochlear implant recipients. Trends in Amplification. 2007;11:193–205. doi: 10.1177/1084713807301379. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Kaernbach C. Simple adaptive testing with the weighted up-down method. Perception & Psychophysics. 1991;49:227–229. doi: 10.3758/bf03214307. [DOI] [PubMed] [Google Scholar]
  6. Miller JD, Dalby JM, Watson CS, Burleson DF. Training experienced hearing-aid users to identify syllable-initial consonants in quiet and noise. J Acoust Soc Am. 2004;115:2387. [Google Scholar]
  7. Miller JD, Dalby JM, Watson CS, Burleson DF. Training experienced hearing-aid users to identify syllable constituents in quiet and noise. ISCA Workshop on Plasticity in Speech Perception, PSP; 2005; June; London. 2005. p. A46. [Google Scholar]
  8. Miller JD, Watson CS, Kewley-Port D, Sillings R, Mills WB, Burleson DF. SPATS: Speech Perception Assessment and Training System. J Acoust Soc Am. 2007;122:3063. doi: 10.1121/1.2942927. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Stecker GC, Bowman GA, Yund EW, Herron TJ, Roup CM, Woods DL. Perceptual training improves syllable identification in new and experienced hearing aid users. J Rehabilitative Res and Dev. 2006;43:537–552. doi: 10.1682/jrrd.2005.11.0171. [DOI] [PubMed] [Google Scholar]
  10. Sweetow R, Henderson-Sabes J. The case for LACE: Listening and auditory communication enhancement training. Hearing Journal. 2004;57:32–38. [Google Scholar]
  11. Sweetow R, Palmer CV. Efficacy of individual auditory training in adults: A systematic review of the evidence. J Am Acad Audiol. 2005;16:494–504. doi: 10.3766/jaaa.16.7.9. [DOI] [PubMed] [Google Scholar]
  12. Sweetow RW, Sabes JH. The need for and development of an adaptive Listening and Communication Enhancement (LACE) Program. J Am Acad Audiol. 2006;17:538–558. doi: 10.3766/jaaa.17.8.2. [DOI] [PubMed] [Google Scholar]

RESOURCES