Born in England, Arthur Boothroyd obtained a degree in Physics from the University of Kingston on Hull in 1957. After a year in electronics research and four ...
Slide 1
Lowest Acceptable Performance Level Arthur Boothroyd and Allison Schauer Presentation to International Collegium of Rehabilitative Audiology October 2015
The results reported here were collected by Allison Schauer during her AuD doctoral research project under the supervision of Dr. Boothroyd. Slide 2
What is the lowest speech Question perception performance consistent with sustained listening? 1. Auditory training? 2. Hearing‐aid processing? 3. Outcome assessment?
This study was prompted by a simple question. What is the lowest level of speech‐perception performance at which at person is likely to sustain listening? The answer has relevance to such issues as auditory training, signal processing in hearing aids, and assessment of the outcome of aural rehabilitation. 1
Slide 3
AUD research project Stephanie Bigler SDSU-UCSD 2012
Effects of computer‐based auditory training on speech perception by adults with hearing loss
This question arose during a recent auditory‐training study. Slide 4
Auditory Training (AudioCasper)
Tester/ scorer
Trainee
Short story, one sentence at a time Variable: SNR, talker speed, “free” words Story context accumulates on screen
We used an adaptive procedure to keep the performance level low enough to promote learning but high enough to sustain participation . But what should that level be? Lacking evidence on this topic, we selected 85% word recognition in the story material used for training. Slide 5
2
Recognition of words in sentences
Context effect k = 7
Residual Noise Penalty at 85% = 16 dB
Quiet
One outcome measure was word recognition in CUNY sentence sets which are designed to provide syntactic, semantic, and topical context. This listener was obviously taking great advantage of context to reach a performance level, in quiet of 90% words recognized. There was weak evidence of a small improvement after training. Even then, this listener required a signal‐ to‐noise ratio some 16 dB higher than young normals to attain a word recognition score in the region of 85%. Slide 6
Present study Falling performance
Increasing challenges Hearing deficit
Speech level & clarity
Noise level & content
Fatigue Distraction Language Etc. complexity
The present study was designed to determine the lowest acceptable performance level. Many factors conspire to reduce speech perception performance. Examples are: • Hearing loss • The level and clarity of the speech • The level and content of interfering noise • Listener fatigue and distraction • The complexity of the language • The importance of the message to the listener ‐and so on. As these challenges increase, performance falls Slide 7 3
Speech perception Falling performance
Acceptable
?
Unacceptable
Increasing challenges
Lowest Acceptable Performance Level
We assume here is a level below which a listener will give up or tune out. This is what we refer to as the lowest acceptable performance level. Slide 8
Participants: 28 young adults with normal hearing
To determine this level, we enlisted 28 young adults listeners – mostly undergraduates in communication disorders.
4
Slide 9
Step 1. Performance vs SNR Percent words recognized in everyday sentences (CUNY) presented at SNR (babble) from -3 to +15 dB
The participants listened to City University of New York (CUNY) everyday sentences in the presence of multi‐talker babble. The sentences were presented at a fixed level of 65 dB HL. Multi‐talker babble was presented at the SNRs of +15, +9, +6, +3, 0, and ‐3. Each participant heard 12 sentences at each SNR. After each sentence, the tester indicated which words were correctly repeated. Slide 10
Performance results ~10 dB
Group mean data for word recognition as a function of SNR were fit with a cubed exponential function. The results indicated that performance rose from 0 at an SNR of around ‐5dB and approached 100% at an SNR of around +6 db – a range of 10 dB. 5
Slide 11
Step 2. Self-assessment 1. Performance (% words recognized)
After testing at each SNR, the listeners were asked to provide three self‐estimates. The first was an estimate of the percentage of words correctly repeated. Slide 12
Hypothetical listening situation
For the next two assessments, listeners were presented with a hypothetical listening situation.
6
Slide 13
Hypothetical listening situation Imagine you are in a restaurant with a group of friends. They are discussing interesting happenings at an event you could not attend. Assuming your level of understanding is equal to what you just experienced, think about your ability and willingness to sustain attention to the conversation
Several features of the situation are important. • The physical context was a restaurant – which is typically noisy and often involves considerable talker‐listener distance • This is a group setting – a situation in which distance can be high and lipreading is often difficult. • The topic is assumed to be of interest to the listener. • The listening difficulty is similar to that just experienced. • Ability and willingness are to be considered separately. • Finally, both are to be assessed in terms of the estimated duration of sustained attention. Slide 14
Self-assessment 1. PERFORMANCE (% words recognized)
2. ABILITY to sustain attention
3. WILLINGNESS to sustain attention
So now we have three self‐estimates: 1. Performance. 2. The length of time the listener believes he/she is ABLE to sustain attention. 3. The length of time the listener believes he/she would be willing to sustain attention. All are understood to be under a condition similar to that just experienced. Listeners made their ability and willingness judgements on a quasi‐logarithmic, categorical, scale of duration. The testing software, however, recorded their responses along an underlying scale of 0 to 100%. 7
Slide 15 Results
Here we see the measured performance data seen earlier. Slide 16 Results
As a group, these listeners underestimated their performance.
8
Slide 17 Results
Here we see the group mean estimate of ability to sustain participation in a group conversation. Slide 18 Results
30 15 5
2 Min‐ utes
Willingness was estimated slightly below willingness Slde 19
9
Slide 20 Results
30 15 5
Lowest acceptable performance level for 5 minutes is ~ 70%
2 Minutes
For now I will focus on measured performance and willingness to sustain listening. A willingness of 5 minutes corresponds with a rating of 40 %. For this group, that rating was reached for an SNR of negative 2dB, corresponding with a word recognition score of around 70% Slide 21 Results
30 15 5
Lowest acceptable performance level for 15 minutes is ~ 90%
Min‐ utes
For 15 minutes of listening, however, the performance level is around 90%, which requires and SNR of 0dB.
10
Slide 22 Results
30 15 5
Lowest acceptable performance level for 30 minutes is ~ 99%
Min‐ utes
And for 30 minutes of listening, the performance level is around 99%, requiring an SNR of +6dB. In other words, the lowest acceptable performance level is not a fixed quantity but varies with the required duration of sustained attention. Slide 23
Individual differences of performance?
It was also clear from these data that individuals differed. This was true for measured performance at lower SNRs.
11
Slide 24
Individual differences of willingness? 30 15 5
2 Minutes
It was even more true for the willingness ratings. Here we see one subject willing to listen for 30 minutes at an SNR of ‐3 dB. And another who was only willing to listen for 5 minutes at an SNR of +15 dB Slide 25
Noise-tolerance domains
Loudness
Annoyance
Distraction
Interference
To assess whether participants were conducting their ratings based on their reaction to the NOISE (rather than the speech), we looked at their reaction to the noise under 4 noise tolerance domains: • Loudness • Annoyance • Distraction • And Interference The choice of domains was based on published research.
12
Slide 26
Noise-tolerance domains
Loudness
Annoyance
Distraction
Interference
One problem with these domains is that they are not entirely orthogonal. Noise can be annoying because it is too loud. Or it can be distracting because it is annoying. Slide 27
Reducing overlap? Loudness Annoyance I don’t mind the way it The noise bothers me. I sounds and it doesn’t don’t like the sound interfere with what I need regardless of how loud it to hear but it is just too is or how much it Distraction Interference loud. It makes me want to interferes with my I don’t sound makes it hard get mind away the from it The noise listening. and the loudness doesn’t for me to follow what the bother me but it takes my person is saying. It attention away from what interferes with what I am I am doing. It is hard to trying to hear concentrate.
In an attempt to increase independence, we provided listeners with definitions. These definitions remained accessible to listeners as they made their ratings. While making their ratings, subjects heard the CUNY sentences at a Speech‐to‐Noise ratio of negative 1.5 dB.
13
Slide 28
Noise-tolerance domains 1. Paired comparisons
The relative importance of the domains was assessed, first, using paired comparisons. Slide 29
Noise-tolerance domains 2. Absolute ratings
And then by giving absolute ratings. For present purposes, the two scores were combined.
14
Slide 30
Means and Std. Errors
The mean ratings for the four domains are shown here. Loudness was of low importance – which was reassuring because the design was intended to protect from uncomfortable loudness. Annoyance received a somewhat higher rating. But distraction from the speech‐perception task and interference with speech‐perception performance were clearly the most important for this group of listeners. Slide 31
Cluster analysis
To look for differences among the listeners we performed a cluster analysis. Because of the small sample size, this analysis was limited to two clusters. The results were quite striking. The subjects divided themselves into two distinct groups. One gave a high rating to the annoying characteristics of the noise but was not very concerned about speech interference. The other was most concerned about interference but was barely concerned about annoyance.
15
Slide 32
Two clusters Group 1
Speech focused
n=19
Noise focused Group 2 n=9
Here we see individual interference ratings plotted against annoyance ratings. The larger cluster consisted of 19 listeners and appeared to be focused on the effects of noise on speech perception. The smaller cluster consisted of 9 listeners and appeared to be focused on the properties of the noise itself. Slide 33
Performance for two clusters
In terms of performance, the two groups showed some significant but very small differences.
16
Slide 34
Willingness for two clusters
In terms of willingness, however, the speech‐focused group was willing to listen for longer at each SNR – including +15 dB. Slide 35 Mackersie, Baxter, & Lane. Noise-tolerance domains underlying Acceptable Noise Level (in revision). Cluster 1. Annoyance‐ driven (n = 10). ANL = 6 dB Cluster 2. Interference‐ driven (n = 20). ANL = 1 dB
I should add that Mackersie and colleagues obtained similar results in a study on noise‐ tolerance and Acceptable Noise Level. The graph shows paired comparison data the top and absolute ratings below. Cluster analysis generated the same clusters with an interference‐driven group and a smaller annoyance‐drive group. The 10 listeners in the annoyance‐drive group had a mean ANL of 6 dB. The 20 listeners in the interference‐driven group had a mean ANL of 1 dB. In other words there was a 5 dB difference in the signal‐to‐noise ratio at which the two groups would “put up with” noise while listening to the narrative.
17
Slide 36
Willingness for 5 minutes
5 min
Lowest Acceptable Level Speech‐focused ~ 65% Noise‐focused ~ 80%
So now we ask, what happens to the conversion of performance and willingness data to acceptable performance if we examine our two groups separately? For 5 minutes of listening, the speech‐focused group required 65% word recognition which was obtained at an SNR of around ‐2dB. But the noise‐focused group required 80% word recognition, which was obtained at an SNR of ‐ 1 dB. Slide 37
Willingness for 15 minutes
15 min
Lowest Acceptable Level Speech‐focused ~ 90% Noise‐focused ~ 98%
For 15 minutes of listening, the lowest acceptable performance was 90% for the speech‐focused group and 98% for the noise‐focused.
18
Slide 38
Conclusion Lowest Acceptable Performance Level: 1. Is not fixed but depends on: a) Duration of listening b) Noise tolerance profile c) Other factors not studied here
We conclude then that the Lowest Acceptable Performance Level is not a fixed quantity but depends on the duration of listening, the noise tolerance of the listener, and probably on many other variables not studied here. Slide 39
Lowest Acceptable Performance Level (in percent words recognized)
Duration Listener focus of listening Speech Noise in minutes dBSNR % dBSNR % 5 ‐2 65 ‐1 80 15 0 90 +3 98 20 +2 95 +7 99 30 +4 99 Not willing
Here is a summary of the findings. For the speech‐focused group, the lowest acceptable performance level rose from 65% to 99% as listening duration rose from 5 to 30 minutes. The SNRs for these performance levels rose from ‐2 to +4 dB. For the noise‐focused group, the lowest acceptable performance level rose from 80% to 99% as listening duration rose from 5 to 20 minutes. The SNRs for these performance levels rose from ‐ 1 to +7 dB. This group was not prepared to sustain listening for 30 minutes. Finally, I want you to note that the SNRs for 20 minutes of listening differed, for the two groups by 5 dB. This value is the same as that found by Mackersie and Colleagues for Acceptable Noise Level in a different subject sample. Slide 40 19
Cautions 1. Self-report 2. By young normal 3. Listening in noise 4. No speechreading 5. Importance/interest 6. Personality
And now a disclaimer. Any attempt to relate these data to hearing‐aid or aural‐rehabilitation research should be conditioned by the following cautions. First, the willingness data were self‐estimates. No attempt was made to find out how long listeners would actually sustain listening under the various conditions tested. Second, these were young normally hearing listeners. Similar data from older listeners with hearing loss would be very interesting. Third, many factors can reduce performance. The only one studied here was noise. Fourth, real‐life situations often allow lipreading to offset some of the effects of diminished hearing. Fifth, the importance and interest of the material to the listener is probably very important. Finally, although we did examine noise tolerance, there are probably many other aspects of personality that influence the length of time for which an individual will sustain listening under difficult conditions. Slide 41
Thanks for listening
20
