Proceedings of Meetings on Acoustics

Transcription

1 Proceedings of Meetings on Acoustics Volume 19, ICA 2013 Montreal Montreal, Canada 2-7 June 2013 Psychological and Physiological Acoustics Session 2pPPb: Speech. Attention, and Impairment (Poster Session) 2pPPb9. Psychometric effects of adding realism to a speech-in-noise test Virginia Best*, Gitte Keidser, Jorg M. Buchholz and Katrina Freeston *Corresponding author's address: National Acoustic Laboratories, Macquarie University, 2109, NSW, Australia, virginia.best@nal.gov.au The speech reception threshold (SRT) is routinely measured in the laboratory to assess speech understanding in noise, but is often reported to be a poor predictor of performance in real world listening situations. The overall goal of this work is to determine whether introducing realistic aspects to speech tests can better capture individual differences and ultimately produce more relevant performance measures. We examined the psychometric effects of (a) transplanting a standard sentence-in-noise test into a simulated reverberant cafeteria environment, and (b) moving from sentence recall to a new ongoing speech comprehension task. Participants included normal hearers and hearing-impaired listeners (who were tested with and without their hearing aids). SRTs in the cafeteria environment were significantly correlated with standard SRTs, but were poorer overall and more sensitive to hearing loss. The comprehension task, despite having very different demands to sentence recall, produced similar SRTs under these conditions. The benefit of hearing aids was weakly correlated across the two listening environments and the two listening tasks. These manipulations promise to be useful for the creation of realistic laboratory tests that are engaging and challenging, yet controlled enough to be useful for psychophysical experiments. Published by the Acoustical Society of America through the American Institute of Physics 2013 Acoustical Society of America [DOI: / ] Received 22 Jan 2013; published 2 Jun 2013 Proceedings of Meetings on Acoustics, Vol. 19, (2013) Page 1

2 INTRODUCTION The speech reception threshold (SRT) is routinely measured in the laboratory to assess speech understanding in noise. However the SRT is often reported to be a poor predictor of performance in real world listening situations. For example, several studies have noted a mismatch between the benefit of hearing aids or processing schemes as measured in the laboratory and how beneficial they are to users in their everyday listening situations [e.g. Bentler et al., 1993; Walden et al., 2000; Cord et al., 2004; Wu, 2010]. There are a number of ecologically relevant variables that are missing from current laboratory speech tests that may prove to be important [Cord et al., 2007; Jerger, 2009]. Among these are dynamic variations in spatial and level characteristics of the acoustic environment, the presence of reverberation, and coordinated visual information. Moreover, sentence repetition is not reflective of real-world communication, which is ongoing rather than discrete and involves the extraction of meaning. A few previous attempts have been made to conduct clinical sentence-based speech testing under more realistic conditions [Killion et al., 1998; Revit et al., 2002; Compton-Conley et al., 2004; Gifford and Revit, 2010]. Although these studies have reported reasonable success with using more realistic environments (e.g. showing expected benefits of directional microphones), to our knowledge there has been no systematic investigation of whether these environments provide different results to those obtained in simpler noise backgrounds. The current study aimed to examine the psychometric effects of two independent manipulations designed to increase the realism of an SRT test. The first manipulation of interest was to move from a typical anechoic laboratory environment to a more realistic acoustic environment. The second manipulation was to use a more engaging task based on comprehension of running speech rather than sentence repetition. The same set of listeners completed the different conditions so that correlations across the conditions could be examined. The ability of the different conditions to capture changes in performance due to amplification in hearing-impaired listeners was also examined. METHODS Participants Forty six listeners participated. Eighteen of these had normal hearing (NH). The other 28 had bilateral sensorineural hearing impairment (HI) that varied in severity but was roughly symmetric (difference between the ears not greater than 25 db at any audiometric frequency). All participants were paid a small amount to cover their travel costs. The pool of listeners ranged in age from 18 to 83 years, and their four-frequency average hearing loss (4FAHL) varied between 1 and 78 db. Twenty five of the 28 HI listeners were regular hearing aid wearers and participated in the experiment both with and without their own hearing aids. For aided testing, their most common hearing aid program was chosen. There were three experimental conditions in total. All listeners completed the first two conditions (standard and realistic). A subset of 11 NH listeners and 26 HI listeners completed the third condition (comprehension). Note however that two of the HI listeners (both 80 years old) could not complete this condition as they could hear but could not keep up enough to answer the questions. Environment and stimuli Testing took place in a large anechoic chamber containing a three dimensional array of radius 1.8 m mounted with 41 identical loudspeakers. The listener was seated at the center of the array, and the experimenter was seated outside the chamber. The two communicated via an intercom, and the experimenter was able to monitor the listener via webcam to ensure he/she maintained a relatively fixed head position. In the standard and realistic conditions, targets were Bamford-Kowal-Bench sentences spoken by an Australian male talker [Bench et al., 1979]. In the comprehension condition, targets were 2-4 min monologues spoken by the same talker using transcripts taken from the listening comprehension component of the International English Language Testing System (IELTS; Cambridge University Press). In the standard and comprehension conditions, targets were presented from the loudspeaker at 0 azimuth, and the background was diffuse multitalker babble generated by presenting four independent samples of 8-talker speech babble from the four loudspeakers positioned at ±45 and ±135 (Figure 1, left). In the realistic condition, a large reverberant room was simulated using ODEON Proceedings of Meetings on Acoustics, Vol. 19, (2013) Page 2

3 software [Rindel, 2000] and a loudspeaker-based auralization toolbox [LoRA; Favrot and Buchholz, 2010]. The target was placed in this simulated room, in front of the listener at a distance of 2.0 m. The background consisted of seven conversations between pairs of Australian-accented talkers who faced each other and were located at various directions and distances (Figure 1, right). The conversations were also based on IELTS transcripts. In all conditions, the background noise was presented continuously at a fixed level of 65 db SPL. FIGURE 1. Schematic layout of the two environments. The standard environment (left) was an anechoic chamber, with a target (T) located directly in front of the listener (L), and four babble maskers (M) located at ±45 and ±135 azimuth. The realistic environment (right) was a simulated reverberant cafeteria, with a frontal target and seven pairs of speech maskers located around the listener. Procedures In the standard and realistic conditions, four blocks of trials were completed. In the first block, an adaptive procedure was used to estimate the 50% SRT. Three blocks of 32 trials were then completed using fixed target levels corresponding to fixed SNRs at the estimated SRT as well as at 2 db above and below the estimated SRT. Listeners spoke aloud their responses, and the experimenter entered the number of correct morphemes (out of a possible 3-8). Each block took approximately 5 min to complete, for a total testing time of around 40 min. In the comprehension condition, listeners were presented with 12 monologues in total, four at each of three different SNRs. Before each presentation they were given a customized answer sheet containing 10 questions, which they gave written answers to in an on-the-go fashion. Each monologue took approximately 5 min to complete, for a total testing time of around 60 min. For all conditions, percent correct scores at the three fixed SNRs were used to generate a psychometric function. Logistic functions were fit to the data (using the psignifit toolbox for MATLAB), and 50% thresholds and slopes were extracted. RESULTS SRTs were strongly correlated in the standard and realistic conditions (Figure 2 top left; r=0.92, p<0.001). However SRTs were consistently higher in the realistic condition, with all points lying above the diagonal. It appears that the difference between conditions is especially strong for the poorer listeners, as indicated by the gradient of the least-squares fit to the data being greater than one (1.41 db/db). Averaged across listeners within a group, the mean increase in SRT in the realistic condition relative to the standard condition was 1.48 db (NH), 4.17 db (HI unaided), and 3.36 db (HI aided). An examination of SRTs as a function of four-frequency average hearing loss confirmed that realistic SRTs were more sensitive to hearing loss (0.17 db/db compared to 0.10 db/db in the standard condition). SRTs were moderately correlated in the standard and comprehension conditions (Figure 2 top right; r=0.77, p<0.001) with the gradient of the least-squares fit relating the two tasks close to one (0.96). Slopes of individual psychometric functions tended to be shallower in the realistic condition than in the standard condition (0.11 versus 0.13), but there was no clear relationship between individual slopes in the two conditions (r=0.14, p=0.26). Slopes were slightly steeper on average in the comprehension condition than in the standard condition (0.18 versus 0.12), but again there was no clear relationship between individual slopes in the two conditions (r=0.10, p=0.48). Proceedings of Meetings on Acoustics, Vol. 19, (2013) Page 3

4 Hearing aid benefits were calculated for each hearing aid wearer by subtracting unaided SRTs from aided SRTs. A huge range of benefits was observed across individuals, including both positive and negative benefits (or disbenefits). Benefits were correlated between the standard and realistic conditions (Figure 2 bottom left; r=0.68, p<0.001), with a tendency for benefits to be greater in the realistic condition (average benefit across all listeners of 1.77 db versus 0.83 db). On average, the benefit of hearing aids was similar for the standard and comprehension tasks (0.83 versus 0.67 db) and individual benefits were weakly correlated (Figure 2 bottom right; r=0.59, p=0.003). FIGURE 2. TOP ROW: Scatterplots showing individual SRTs in the realistic condition (left) and comprehension condition (right) against SRTs in the standard condition. Different symbols indicate NH listeners (squares), unaided HI listeners (filled circles), and aided HI listeners (open circles). BOTTOM ROW: Scatterplots showing individual hearing aid benefits in the realistic condition (left) and comprehension condition (right) against hearing aid benefits in the standard condition. Solid lines in all panels show least squares fits. DISCUSSION Effect of using a more realistic environment Results from this experiment indicate that the use of a more complex and realistic acoustic environment can change the psychometric properties of a simple sentence test. Thresholds were higher on average in the realistic environment, and slopes tended to be shallower. Shallower slopes were expected in the cafeteria noise due to the increase in acoustic variability [Durlach et al., 2003; Durlach et al., 2005]. Previous studies have demonstrated that intelligible speech maskers can cause informational masking and lead to increased thresholds [Carhart et al., 1969; Brungart, 2001; Brungart et al., 2001; Best et al., 2012]. Reverberation also tends to increase thresholds [Culling et al., 2003; Lavandier and Culling, 2007; 2008]. Further work is required to determine the relative contribution of these two factors (and possibly others) in our study. In general though, modifications to speech tests that shift SRTs towards positive SNRs are useful for the goal of increasing real-world relevance, as environmental SNRs are generally above 0 db [Pearsons et al., 1976; Smeds et al., 2012]. Proceedings of Meetings on Acoustics, Vol. 19, (2013) Page 4

5 The increase in SRTs in the realistic condition was particularly strong for the listeners with the poorest hearing. This may be related to the fluctuating background and/or to the presence of reverberation. Reverberation is a strong candidate given that several previous studies have reported an interaction between reverberation and hearing loss, particularly in noise [Harris and Reitz, 1985; Nabelek, 1988; Harris and Swenson, 1990]. The increase in sensitivity to hearing loss and the overall larger range of SRTs implies that the realistic environment magnified the differences between listeners. This may be useful when the goal is to discriminate normal hearers from hearing-impaired listeners, or to reveal functional differences between individuals with different degrees of hearing loss. Another question of interest was whether the characteristics of the environment would influence the measured benefit of hearing aids for understanding speech in noise. A lot of variability was observed, in part because the listeners were not uniformly fitted with identical devices. Nonetheless, the benefits were correlated across the two environments with a slight tendency for larger benefits in the realistic condition. This may be a corollary of the increased sensitivity to hearing loss (i.e. an increased sensitivity to hearing restoration), or it may be related to the shift towards more positive SNRs (where hearing aids are most effective). It would be of interest to extend this work to examine the benefits provided by directional microphones, whose effectiveness is SNR-sensitive and for which current laboratory measures are rather unsatisfactory [Cord et al., 2004; Cord et al., 2007; Wu, 2010]. Effect of using a comprehension task There has long been interest amongst hearing researchers in speech tasks that more closely resemble real-world speech communication. However the psychophysical consequences of moving to such a task have never been closely investigated. Thus, the second aim of this study was to use our large group of listeners to examine the effects of moving from sentence recall to passage comprehension. We expected to find substantial changes in the psychometric properties, based on the results of other studies of speech comprehension [e.g. Tye-Murray et al., 2008]. However, SRTs were highly correlated across the two tasks in our experiment. The absence of a strong effect may be a result of the on-the-go response format we used, which removed the memory load typical of other comprehension tasks. Moreover, the fact that our test was conducted in noise may mean that audibility of keywords was the primary limitation, effectively minimizing any effects of individual differences in language skills or cognitive ability that may affect comprehension under different conditions. However, most listeners did report that the comprehension task was more challenging than the standard task, and certainly the requirement to switch between listening, reading, processing information, and writing seems intuitively more demanding than sentence repetition. It is worth noting that two of the HI listeners (both 80 years old) could not complete the comprehension task even in quiet, despite reasonable abilities on the sentence test in noise. Thus it is possible that this style of task does have the potential to reveal age-related cognitive limitations that sentence tests may not. CONCLUSIONS The current study aimed to examine the psychometric effects of two independent manipulations designed to increase the realism of an SRT test. The use of a more realistic acoustic environment led to higher SRTs that were somewhat more sensitive to both hearing loss and amplification. The use of speech comprehension did not have dramatic effects on the psychometric properties but was reported to be a more challenging task. Overall, these manipulations promise to be useful for the creation of realistic laboratory tests that are engaging and natural, yet controlled enough to be useful for psychophysical experiments. The consistent correlations we found between conditions suggest that the manipulations examined did not substantially change the overall ranking of listeners; the good listeners in one condition were the good listeners in the other, etc. Thus we would not expect large improvements in our ability to predict individual real-world performance with either of these manipulations. It is of course possible that environment and task characteristics may interact, and ultimately a combination of ecologically motivated modifications might be the way to achieve more relevant outcome measures. ACKNOWLEDGMENTS The authors acknowledge the financial support of the HEARing CRC, established and supported under the Cooperative Research Centres Program, an Australian Government initiative. We are also grateful to Harvey Dillon for several helpful discussions. Proceedings of Meetings on Acoustics, Vol. 19, (2013) Page 5

6 REFERENCES Bench, R. J., Doyle, J., Daly, N., and Lind, C. (1979). The BKB/A speechreading (lipreading) test (La Trobe University, Victoria). Bentler, R. A., Niebuhr, D. P., Getta, J. P., and Anderson, C. V. (1993). "Longitudinal study of hearing aid effectiveness II. Subjective measures," J. Speech Hear. Res. 36, Best, V., Marrone, N., Mason, C. R., and Kidd Jr., G. (2012). "The influence of non-spatial factors on measures of spatial release from masking," J. Acoust. Soc. Am. 13, Brungart, D. S. (2001). "Informational and energetic masking effects in the perception of two simultaneous talkers," J. Acoust. Soc. Am. 109, Brungart, D. S., Simpson, B. D., Ericson, M. A., and Scott, K. R. (2001). "Informational and energetic masking effects in the perception of multiple simultaneous talkers," J. Acoust. Soc. Am. 110, Carhart, R., Tillman, T. W., and Greetis, E. S. (1969). "Perceptual masking in multiple sound backgrounds," J. Acoust. Soc. Am. 45, Compton-Conley, C. L., Neuman, A. C., Killion, M., and Levitt, H. (2004). "Performance of directional microphones for hearing aids: real-world versus simulation," J. Am. Acad. Audiol. 15, Cord, M., Baskent, D., Kalluri, S., and Moore, B. C. J. (2007). "Disparity between clinical assessment and real-world performance of hearing aids," Hear. Rev. Cord, M. T., Surr, R. K., Walden, B. E., and Dyrlund, O. (2004). "Relationship between laboratory measures of directional advantage and everyday success with directional microphone hearing aids," J. Am. Acad. Audiol. 15, Culling, J. F., Hodder, K. I., and Toh, C. Y. (2003). "Effects of reverberation on perceptual segregation of competing voices," J. Acoust. Soc. Am. 114, Durlach, N., Mason, C., Shinn-Cunningham, B., Arbogast, T., Colburn, H., and Kidd Jr, G. (2003). "Informational masking: counteracting the effects of stimulus uncertainty by decreasing target-masker similarity," J. Acoust. Soc. Am. 114, Durlach, N. I., Mason, C. R., Gallun, F. J., Shinn-Cunningham, B. G., Colburn, H. S., and Kidd Jr, G. (2005). "Psychometric functions for fixed and randomly mixed maskers," J. Acoust. Soc. Am. 118, Favrot, S., and Buchholz, J. M. (2010). "LoRA A loudspeaker-based room auralization system," ACUSTICA/acta acustica 96, Gifford, R. H., and Revit, L. J. (2010). "Speech perception for adult cochlear implant recipients in a realistic background noise: Effectiveness of preprocessing strategies and external options for improving speech recognition in noise," J. Am. Acad. Audiol. 21, Harris, R. W., and Reitz, M. L. (1985). "Effects of room reverberation and noise on speech discrimination by the elderly," Audiology 24, Harris, R. W., and Swenson, D. W. (1990). "Effects of reverberation and noise on speech recognition by adults with various amounts of sensorineural hearing impairment," Audiology 29, Jerger, J. (2009). "Ecologically valid measures of hearing aid performance," Starkey Audiology Series 1, 1-4. Killion, M., Schulein, R., Christensen, L., Fabry, D., Revit, L., Niquette, P., and Chung, K. (1998). "Real-world performance of an ITE directional microphone," Hearing Journal 51, 1-6. Lavandier, M., and Culling, J. F. (2007). "Speech segregation in rooms: effects of reverberation on both target and interferer," J. Acoust. Soc. Am. 122, Lavandier, M., and Culling, J. F. (2008). "Speech segregation in rooms: monaural, binaural, and interacting effects of reverberation on target and interferer," J. Acoust. Soc. Am. 123, Nabelek, A. K. (1988). "Identification of vowels in quiet, noise, and reverberation: Relationships with age and hearing loss," J. Acoust. Soc. Am. 84, Pearsons, K. S., Bennett, R. L., and Fidell, S. (1976). "Speech levels in various noise environments," Environmental Health Effects Research Series, Office of Research and Development, U.S. Environmental Protection Agency EPA-600/ Revit, L. J., Schulein, R. B., and Julstrom, S. (2002). "Toward accurate assessment of real-world hearing aid benefit," Hearing Review 9, Rindel, J. H. (2000). "The use of computer modeling in room acoustics," J. Vibroengineering 3, Smeds, K., Wolters, F., and Rung, M. (2012). "Realistic signal-to-noise ratios," IHCON, Lake Tahoe, CA. Tye-Murray, N., Sommers, M., Spehar, B., Myerson, J., Hale, S., and Rose, N. (2008). "Auditory-visual discourse comprehension by older and young adults in favorable and unfavorable conditions," Int. J. Audiol. 47, S31-S37. Walden, B. E., Surr, R. K., Cord, M. T., Edwards, B., and Olson, L. (2000). "Comparison of benefits provided by different hearing aid technologies," J. Am. Acad. Audiol. 11, Wu, T.-H. (2010). "Effect of age on directional microphone hearing aid benefit and preference," J. Am. Acad. Audiol. 21, Proceedings of Meetings on Acoustics, Vol. 19, (2013) Page 6