J. COMMUN. DISORD. 24 (1991). 180-186
AN EVALUATION OF TWO SIGNALPROCESSING HEARING AIDS JAMES J. DEMPSEY
Department of Speech, Communication Sciences, and Thearre, St. John’s Lexington Hearing and Speech Center, Jackson Heights, New York
University,
and
TANYA G. LINZALONE Lexington
Hearing
and Speech
Center.
Jackson
Heights,
New
York
Various forms of signal processing are used in modern ASP hearing aids. The present study investigated the relationship between sentence recognition ability and two types of signal processing used in commercially available hearing aids. Results indicated a significant improvement in sentence recognition ability employing an instrument with adaptive compression (variable release time) versus an instrument with an adaptive high-pass filter with short attack and release times. Data were obtained for a single-talker competing message at several message-to-competition ratios. The adaptive compression system may prove to be beneficial to hearing-impaired listeners in certain background noise environments.
INTRODUCTION One of the most common complaints of hearing-aid users is difftculty understanding speech in noisy situations. Hearing aids with automatic signal processing (ASP) have been designed to address the problem of speech recognition in noise (Dempsey, 1987). These ASP instruments generally attempt to suppress noise through the use of either low-frequency attenuation or some other specific-frequency attenuation (Tyler and Kuk, 1989). In addition to some type of frequency-shaping feature, the typical ASP hearing aid uses an input amplitude compression system in an attempt to maintain a constant most comfortable listening level (MCL) (Griffing and Heide, 1983). Of primary concern when employing a compression system is the attack and release time of the compressing amplifier. The attack time has little effect on the received signal when kept under 10 msec, whereas the length of the release time can affect the signal-to-noise ratio produced by the instrument (Smriga, 1987). The signal-to-noise ratio is critically related to speech recognition ability. Since Address of Speech
Correspondence Communication
180 0021.9924/91/$3.50
to James J. Dempsey, Ph.D., St. John’s University. Sciences, and Theatre, Jamaica, NY 11439. 0 195’1by Elsevier 655 Avenue
Department
Science Publishing Co.. Inc. of the Americas, New York, NY 10010
RELEASE TIME AND SENTENCE RECOGNITION
181
the aim of these signal-processing modifications is to improve the reception of acoustic speech cues by the hearing-impaired listener, it becomes important to investigate the relationship between speech recognition ability and the release time of a signal-processing system. The purpose of the present investigation, therefore, was to compare speech recognition scores obtained by hearing-impaired listeners in backgrounds of noise using both a signal-processing hearing aid with a variable release time and a signal-processing hearing aid that employs an instantaneous, high-pass filter with virtually no release time. METHOD Subjects Fifteen adults with moderate to severe, bilaterally symmetrical sensorineural hearing losses participated in this study. The mean age of the subjects was 67 years, with a range of 62-77 years. All subjects were experienced hearing-aid users. Instrumentation Two commercially available hearing aids were utilized. The instruments were a Telex 363 C behind-the-ear (BTE) and an Argosy, stock in-theear (ITE) with the Manhattan I Circuit. These instruments were selected on the basis of having similar frequency response and gain characteristics but different methods of signal processing. The Telex 363 employs an adaptive compression system which varies the release time of the compression circuitry from 50 to 550 msec based upon the amplifier input. A short release time is used when the noise is sporadic and short in duration whereas the release time is increased for noises that are continual and relatively steady state. The Argosy with the Manhattan Circuit was selected as a comparison instrument because the noise-reduction circuit operates as an adaptive high-pass filter with relatively short attack and release times. One of the instruments was randomly selected as the reference instrument for each subject. The reference instrument was then set to a most comfortable listening level (MCL) in quiet and the output of the instruments was equalized within t2 dB for the frequencies 5003000 Hz. This output matching allowed for a comparison of adaptive compression versus high-pass filtering that was not contaminated by the user’s gain setting. Procedures The subjects were seated between two loudspeakers in a sound-treated audiological test booth. Sentence recognition ability was assessed using audiotape recordings of the Synthetic Sentence Identification Test (SSI).
J. DEMPSEY and T. LINZALONE
182
Table 1. Average SSI Scores in Percent Correct Hearing Aid; and Each Message-to-Competition
for All Subjects; Ratio”
Message-to-competition -20 dB Adaptive compression Adaptive high-pass filtering
6.7
- 10 dB
Each ratio
0 dB
+lOdB 76.0
23.7
43.7
(1.9)
(3.7)
(6.7)
(7.9)
13.7 (3.1)
39.7 (4.3)
62.7 (8.1)
87.0 (4.9)
” Standard errcm in parentheses.
The competing message consisted of a continuous single-talker background noise. Scores were obtained for each of the two hearing aids and four background noise conditions (- 20, - 10, 0, and + 10 dB messageto-competition ratios). A counterbalanced test order across subjects was used for the sentence order and background noise levels. A total of eight sentence recognition scores was obtained from each subject. The target sentences originated from 0” azimuth at an intensity level of 70 dB SPL. The competing message was presented from 180” azimuth at 60, 70, 80, or 90 dB SPL. The subject was aided monaurally, with the test ear being the ear in which the subject normally wore a hearing aid. An EAR hearing protector was placed in the unaided ear during all testing. These plugs provide attenuation of approximately 30-35 dB for the frequencies 500-8000 Hz (Humes, 1983).
RESULTS Table 1 contains the mean and standard error of the mean for sentence recognition scores obtained for all subjects with each hearing aid and each background noise condition. Inspection of Table I reveals superior sentence recognition scores with adaptive compression versus adaptive highpass filtering for all conditions, The greatest average differences were observed at message-to-competition ratios of - 10 and 0 dB. A repeated measures analysis of variance (ANOVA) was used to test the statistical significance of these sentence recognition scores (Madigan and Lawrence, 1983). The experimental design consisted of two aided listening conditions (adaptive compression versus adaptive high-pass filtering) tested at four background noise levels (- 20, - 10, 0, and + 10 dB message-to-competition ratios). The ANOVA revealed significant 0, < 0.001) main effects for both the hearing aid factor and the background noise factor. The interaction of instrument and noise level was not significant (p > 0.1).
RELEASE
TIME AND SENTENCE
RECOGNITION
183
10
0 12
3
4
5
6
7
8
9
10
11
12
13
14
15
SUBJECT Figure 1. Percent
improvement in sentence recognition provided by the Telex (adaptive compression) hearing aid versus the Argosy (Manhattan Circuit) hearing aid for each subject averaged across the four message-to-competition ratios.
4ob
1234567
8
9
10
11
12
13
14
15
SUBJECT Figure 2. Percent
improvement in sentence recognition provided by the Telex (adaptive compression) hearing aid versus the Argosy (Manhattan Circuit) hearing aid for each subject with a message-to-competition ratio of - 10 dB. Subjects 3, 12, and 14 demonstrated no difference between instruments for this condition.
J. DEMPSEY and T. LINZALONE
50
40
30
20
10
0 12345676
9
10
11
12
13
14
15
SUBJECT
Percent improvement in sentence recognition provided by the Telex (adaptive compression) hearing aid versus the Argosy (Manahattan Circuit) hearing aid for each subject with a message-to-competition ratio of 0 dB. Subject 14 demonstrated no difference between instruments for this condition.
Figure 3.
Figure 1 contains the average advantage (difference scores) provided by the adaptive compression compared with adaptive high-pass filtering for each subject collapsed over all four message-to-competition ratios. Eight out of the 15 subjects demonstrated improvements in sentence recognition of 10% or better. The most striking advantages of the adaptive compression system over adaptive high-pass filtering were obtained with message-to-competition ratios of - 10 and 0 dB. Figures 2 and 3 contain the advantage (difference scores) provided by the adaptive compression circuit versus adaptive high-pass filtering for each subject with message-to-competition ratios of - 10 and 0 dB, respectively. Inspection of these figures reveals improvements of 20 percentage points or more for 8 out of 15 subjects for each background noise level. A subjective evaluation of the two instruments revealed a preference for adaptive high-pass filtering. Of the 15 subjects in the present investigation, nine subjects rated the hearing aid with adaptive high-pass filtering as having better sound quality and stated that they would prefer wearing it over the instrument with adaptive compression.
RELEASE TIME AND SENTENCE RECOGNITION
185
DISCUSSION The subjects in the present investigation demonstrated superior sentence recognition scores with the adaptive compression system versus adaptive high-pass filtering. The adaptive compression system provides a long release time (up to 550 msec) in long-duration, high-intensity noise background such as that employed in this study. The adaptive, high-frequency filter had a relatively small release time. The instruments were matched for frequency response and gain characteristics in quiet, suggesting that for the type of speech interference considered in this study, adaptive compression is superior to adaptive high-pass filtering in improving sentence recognition scores. A single-speaker competing message has been used previously in evaluating automatic signal-processing hearing aids (Stach, Speerschneider, and Jerger, 1987). This type of background noise was selected because it is particularly damaging to intelligibility. It is also the type of interference that is particularly difficult to filter out because both the desired speech signal and the competing speech have similar average spectra. Adaptive high-pass filtering is best suited for combating low-frequency ambient noise, such as that produced by many home appliances. Adaptive compression, on the other hand, is well suited for handling time-varying speech and noise spectra, as shown by the results of this study. Although this study focused on a form of interference for which adaptive compression is well suited, it is significant to note that the majority of subjects preferred the sound quality of adaptive high-pass filtering. This finding is consistent with that reported by Sigelman and Preves (1987). The results of the current study should not be generalized too broadly. It is to be anticipated that different methods of signal processing will offer different advantages for different listening conditions. This study focused on one of the more common listening situations that hearing-aid users find particularly troublesome, that of a single, competing speaker. Other common types of interference may well show different results. Another important consideration is that of candidacy. Persons with severe recruitment, for example, are likely to benefit from advanced forms of compression amplification (Villchur, 1973). New hearing-aid users are also more likely to differ in their preferences and ratings of relative sound quality from long-term hearing-aid users. It is unlikely that any single form of signal processing will prove to be superior for all listening situations or all hearing-aid candidates. It is, nevertheless, important to document the relative advantages of the different methods of signal processing. This information will not only facilitate the selection of the most appropriate instrument for each hearing-aid candidate, but will also provide the data needed to develop improved signal-processing hearing aids.
186
J. DEMPSEY
and T. LINZALONE
CONCLUSIONS In summary, sentence recognition scores with a single-speaker competing message were significantly better using a hearing aid with adaptive compression (variable release time) compared to a hearing aid with adaptive high-pass filtering. The clinical implication is that adaptive compression appears to be well suited for those hearing-impaired individuals with complaints regarding understanding speech when more than one person is talking. Further evaluation is necessary to determine if this advantage is maintained in other background noise environments. This study was supported in part by the National Institute of Disability and Rehabilitation Research, under grant No. 133E80019.
REFERENCES Dempsey, J. (1987). Effect of automatic signal-processing amplification on speech recognition in noise for persons with sensorineural hearing loss. Ann. Otol. Rhinol.
Laryngol.
96:251-253.
Griffing, T., and Heide, J. (1983). Automatic definition. Hear Instrum. 3414.
signal processor
Humes, L. (1983). A psychophysical evaluation protector attenuation on noise level. J. Acoust. Madigan, S., and Lawrence, ance Program
for the Apple
aids. Part I: By
of the dependence Sot.
of hearing
Am. 73:297-311.
V. (1983). ANOVA II: A General Analysis of VariII. Northridge, CA: Human Systems Dynamics.
Sigelman, J., and Preves, D. (1987). Field trials of a new adaptive signal processor hearing aid circuit. Hear. .I. 40:24-29. Smriga, D. (1987). Improving speech perception in noise: The frequency and the intensity domain. Hear. Znstrum. 38:8. Stach, B., Speerschneider, J., and Jerger, J. (1987). Evaluating automatic signal processing hearing aids. Hear. J. 40: 15-19.
domain
the efficacy of
Tyler, R., and Kuk, F. (1989). The effects of “noise suppression” hearing aid on consonant recognition in speech-babble and low-frequency noise. Ear Hear. 10:243-249.
Villchur, E. (1973). Signal processing to improve speech intelligibility tive deafness. J. Acoust. Sot. Am. 53: 1646-1657.
in percep-
AN EVALUATION OF TWO SIGNALPROCESSING HEARING AIDS JAMES J. DEMPSEY
Department of Speech, Communication Sciences, and Thearre, St. John’s Lexington Hearing and Speech Center, Jackson Heights, New York
University,
and
TANYA G. LINZALONE Lexington
Hearing
and Speech
Center.
Jackson
Heights,
New
York
Various forms of signal processing are used in modern ASP hearing aids. The present study investigated the relationship between sentence recognition ability and two types of signal processing used in commercially available hearing aids. Results indicated a significant improvement in sentence recognition ability employing an instrument with adaptive compression (variable release time) versus an instrument with an adaptive high-pass filter with short attack and release times. Data were obtained for a single-talker competing message at several message-to-competition ratios. The adaptive compression system may prove to be beneficial to hearing-impaired listeners in certain background noise environments.
INTRODUCTION One of the most common complaints of hearing-aid users is difftculty understanding speech in noisy situations. Hearing aids with automatic signal processing (ASP) have been designed to address the problem of speech recognition in noise (Dempsey, 1987). These ASP instruments generally attempt to suppress noise through the use of either low-frequency attenuation or some other specific-frequency attenuation (Tyler and Kuk, 1989). In addition to some type of frequency-shaping feature, the typical ASP hearing aid uses an input amplitude compression system in an attempt to maintain a constant most comfortable listening level (MCL) (Griffing and Heide, 1983). Of primary concern when employing a compression system is the attack and release time of the compressing amplifier. The attack time has little effect on the received signal when kept under 10 msec, whereas the length of the release time can affect the signal-to-noise ratio produced by the instrument (Smriga, 1987). The signal-to-noise ratio is critically related to speech recognition ability. Since Address of Speech
Correspondence Communication
180 0021.9924/91/$3.50
to James J. Dempsey, Ph.D., St. John’s University. Sciences, and Theatre, Jamaica, NY 11439. 0 195’1by Elsevier 655 Avenue
Department
Science Publishing Co.. Inc. of the Americas, New York, NY 10010
RELEASE TIME AND SENTENCE RECOGNITION
181
the aim of these signal-processing modifications is to improve the reception of acoustic speech cues by the hearing-impaired listener, it becomes important to investigate the relationship between speech recognition ability and the release time of a signal-processing system. The purpose of the present investigation, therefore, was to compare speech recognition scores obtained by hearing-impaired listeners in backgrounds of noise using both a signal-processing hearing aid with a variable release time and a signal-processing hearing aid that employs an instantaneous, high-pass filter with virtually no release time. METHOD Subjects Fifteen adults with moderate to severe, bilaterally symmetrical sensorineural hearing losses participated in this study. The mean age of the subjects was 67 years, with a range of 62-77 years. All subjects were experienced hearing-aid users. Instrumentation Two commercially available hearing aids were utilized. The instruments were a Telex 363 C behind-the-ear (BTE) and an Argosy, stock in-theear (ITE) with the Manhattan I Circuit. These instruments were selected on the basis of having similar frequency response and gain characteristics but different methods of signal processing. The Telex 363 employs an adaptive compression system which varies the release time of the compression circuitry from 50 to 550 msec based upon the amplifier input. A short release time is used when the noise is sporadic and short in duration whereas the release time is increased for noises that are continual and relatively steady state. The Argosy with the Manhattan Circuit was selected as a comparison instrument because the noise-reduction circuit operates as an adaptive high-pass filter with relatively short attack and release times. One of the instruments was randomly selected as the reference instrument for each subject. The reference instrument was then set to a most comfortable listening level (MCL) in quiet and the output of the instruments was equalized within t2 dB for the frequencies 5003000 Hz. This output matching allowed for a comparison of adaptive compression versus high-pass filtering that was not contaminated by the user’s gain setting. Procedures The subjects were seated between two loudspeakers in a sound-treated audiological test booth. Sentence recognition ability was assessed using audiotape recordings of the Synthetic Sentence Identification Test (SSI).
J. DEMPSEY and T. LINZALONE
182
Table 1. Average SSI Scores in Percent Correct Hearing Aid; and Each Message-to-Competition
for All Subjects; Ratio”
Message-to-competition -20 dB Adaptive compression Adaptive high-pass filtering
6.7
- 10 dB
Each ratio
0 dB
+lOdB 76.0
23.7
43.7
(1.9)
(3.7)
(6.7)
(7.9)
13.7 (3.1)
39.7 (4.3)
62.7 (8.1)
87.0 (4.9)
” Standard errcm in parentheses.
The competing message consisted of a continuous single-talker background noise. Scores were obtained for each of the two hearing aids and four background noise conditions (- 20, - 10, 0, and + 10 dB messageto-competition ratios). A counterbalanced test order across subjects was used for the sentence order and background noise levels. A total of eight sentence recognition scores was obtained from each subject. The target sentences originated from 0” azimuth at an intensity level of 70 dB SPL. The competing message was presented from 180” azimuth at 60, 70, 80, or 90 dB SPL. The subject was aided monaurally, with the test ear being the ear in which the subject normally wore a hearing aid. An EAR hearing protector was placed in the unaided ear during all testing. These plugs provide attenuation of approximately 30-35 dB for the frequencies 500-8000 Hz (Humes, 1983).
RESULTS Table 1 contains the mean and standard error of the mean for sentence recognition scores obtained for all subjects with each hearing aid and each background noise condition. Inspection of Table I reveals superior sentence recognition scores with adaptive compression versus adaptive highpass filtering for all conditions, The greatest average differences were observed at message-to-competition ratios of - 10 and 0 dB. A repeated measures analysis of variance (ANOVA) was used to test the statistical significance of these sentence recognition scores (Madigan and Lawrence, 1983). The experimental design consisted of two aided listening conditions (adaptive compression versus adaptive high-pass filtering) tested at four background noise levels (- 20, - 10, 0, and + 10 dB message-to-competition ratios). The ANOVA revealed significant 0, < 0.001) main effects for both the hearing aid factor and the background noise factor. The interaction of instrument and noise level was not significant (p > 0.1).
RELEASE
TIME AND SENTENCE
RECOGNITION
183
10
0 12
3
4
5
6
7
8
9
10
11
12
13
14
15
SUBJECT Figure 1. Percent
improvement in sentence recognition provided by the Telex (adaptive compression) hearing aid versus the Argosy (Manhattan Circuit) hearing aid for each subject averaged across the four message-to-competition ratios.
4ob
1234567
8
9
10
11
12
13
14
15
SUBJECT Figure 2. Percent
improvement in sentence recognition provided by the Telex (adaptive compression) hearing aid versus the Argosy (Manhattan Circuit) hearing aid for each subject with a message-to-competition ratio of - 10 dB. Subjects 3, 12, and 14 demonstrated no difference between instruments for this condition.
J. DEMPSEY and T. LINZALONE
50
40
30
20
10
0 12345676
9
10
11
12
13
14
15
SUBJECT
Percent improvement in sentence recognition provided by the Telex (adaptive compression) hearing aid versus the Argosy (Manahattan Circuit) hearing aid for each subject with a message-to-competition ratio of 0 dB. Subject 14 demonstrated no difference between instruments for this condition.
Figure 3.
Figure 1 contains the average advantage (difference scores) provided by the adaptive compression compared with adaptive high-pass filtering for each subject collapsed over all four message-to-competition ratios. Eight out of the 15 subjects demonstrated improvements in sentence recognition of 10% or better. The most striking advantages of the adaptive compression system over adaptive high-pass filtering were obtained with message-to-competition ratios of - 10 and 0 dB. Figures 2 and 3 contain the advantage (difference scores) provided by the adaptive compression circuit versus adaptive high-pass filtering for each subject with message-to-competition ratios of - 10 and 0 dB, respectively. Inspection of these figures reveals improvements of 20 percentage points or more for 8 out of 15 subjects for each background noise level. A subjective evaluation of the two instruments revealed a preference for adaptive high-pass filtering. Of the 15 subjects in the present investigation, nine subjects rated the hearing aid with adaptive high-pass filtering as having better sound quality and stated that they would prefer wearing it over the instrument with adaptive compression.
RELEASE TIME AND SENTENCE RECOGNITION
185
DISCUSSION The subjects in the present investigation demonstrated superior sentence recognition scores with the adaptive compression system versus adaptive high-pass filtering. The adaptive compression system provides a long release time (up to 550 msec) in long-duration, high-intensity noise background such as that employed in this study. The adaptive, high-frequency filter had a relatively small release time. The instruments were matched for frequency response and gain characteristics in quiet, suggesting that for the type of speech interference considered in this study, adaptive compression is superior to adaptive high-pass filtering in improving sentence recognition scores. A single-speaker competing message has been used previously in evaluating automatic signal-processing hearing aids (Stach, Speerschneider, and Jerger, 1987). This type of background noise was selected because it is particularly damaging to intelligibility. It is also the type of interference that is particularly difficult to filter out because both the desired speech signal and the competing speech have similar average spectra. Adaptive high-pass filtering is best suited for combating low-frequency ambient noise, such as that produced by many home appliances. Adaptive compression, on the other hand, is well suited for handling time-varying speech and noise spectra, as shown by the results of this study. Although this study focused on a form of interference for which adaptive compression is well suited, it is significant to note that the majority of subjects preferred the sound quality of adaptive high-pass filtering. This finding is consistent with that reported by Sigelman and Preves (1987). The results of the current study should not be generalized too broadly. It is to be anticipated that different methods of signal processing will offer different advantages for different listening conditions. This study focused on one of the more common listening situations that hearing-aid users find particularly troublesome, that of a single, competing speaker. Other common types of interference may well show different results. Another important consideration is that of candidacy. Persons with severe recruitment, for example, are likely to benefit from advanced forms of compression amplification (Villchur, 1973). New hearing-aid users are also more likely to differ in their preferences and ratings of relative sound quality from long-term hearing-aid users. It is unlikely that any single form of signal processing will prove to be superior for all listening situations or all hearing-aid candidates. It is, nevertheless, important to document the relative advantages of the different methods of signal processing. This information will not only facilitate the selection of the most appropriate instrument for each hearing-aid candidate, but will also provide the data needed to develop improved signal-processing hearing aids.
186
J. DEMPSEY
and T. LINZALONE
CONCLUSIONS In summary, sentence recognition scores with a single-speaker competing message were significantly better using a hearing aid with adaptive compression (variable release time) compared to a hearing aid with adaptive high-pass filtering. The clinical implication is that adaptive compression appears to be well suited for those hearing-impaired individuals with complaints regarding understanding speech when more than one person is talking. Further evaluation is necessary to determine if this advantage is maintained in other background noise environments. This study was supported in part by the National Institute of Disability and Rehabilitation Research, under grant No. 133E80019.
REFERENCES Dempsey, J. (1987). Effect of automatic signal-processing amplification on speech recognition in noise for persons with sensorineural hearing loss. Ann. Otol. Rhinol.
Laryngol.
96:251-253.
Griffing, T., and Heide, J. (1983). Automatic definition. Hear Instrum. 3414.
signal processor
Humes, L. (1983). A psychophysical evaluation protector attenuation on noise level. J. Acoust. Madigan, S., and Lawrence, ance Program
for the Apple
aids. Part I: By
of the dependence Sot.
of hearing
Am. 73:297-311.
V. (1983). ANOVA II: A General Analysis of VariII. Northridge, CA: Human Systems Dynamics.
Sigelman, J., and Preves, D. (1987). Field trials of a new adaptive signal processor hearing aid circuit. Hear. .I. 40:24-29. Smriga, D. (1987). Improving speech perception in noise: The frequency and the intensity domain. Hear. Znstrum. 38:8. Stach, B., Speerschneider, J., and Jerger, J. (1987). Evaluating automatic signal processing hearing aids. Hear. J. 40: 15-19.
domain
the efficacy of
Tyler, R., and Kuk, F. (1989). The effects of “noise suppression” hearing aid on consonant recognition in speech-babble and low-frequency noise. Ear Hear. 10:243-249.
Villchur, E. (1973). Signal processing to improve speech intelligibility tive deafness. J. Acoust. Sot. Am. 53: 1646-1657.
in percep-
