The objective of this study was to establish guidelines for specifying the acceptable output characteristics of assistive listening devices for people with hearing loss. Fifty-nine adult listeners (49 with hearing loss and 10 without) listened binaurally to sentence materials that were subjected to three different types of distortion; reverberation and background noise, internally-generated induction loop noise, and peak clipping. The listeners provided ratings as to the quality of the materials presented. A minimally acceptable criterion was selected and results for the listeners with hearing loss were compared with that criterion to arrive at:
- Minimally acceptable output and dynamic range levels,
- Minimally acceptable Speech Transmission Index (STI) level,
- Minimally acceptable signal-to-noise ratio for internally generated noise, and
- Minimally acceptable peak clipping level.
Forty-nine adult listeners with sensorineural hearing loss participated in this study. They were divided into six groups according the degree and configuration of their hearing loss. The following is a list of criteria for inclusion in each group.
- Very high frequency hearing loss: the three-frequency pure tone average (PTA) is less than 41 dB HL with thresholds at 2000 Hz less than 40 dB HL.
- Moderate flat hearing loss: PTA is greater than 41 dB HL with threshold at 2000 Hz less than 55 dB HL.
- Moderately-severe, gently sloping hearing loss: PTA is greater than 41 dB HL with threshold at 2000 Hz between 55 and 60 dB HL).
- Moderate to severe flat hearing loss: PTA is greater than 41 dB HL with thresholds at 2000 Hz between 61 and 70 dB HL.
- Moderate to severe sloping hearing loss: PTA is greater than 41 dB HL with thresholds at 2000 Hz greater than 70 dB HL.
- Precipitous, high- frequency hearing loss: PTA is less than 41 d B HL with thresholds at 2000 Hz greater than 40 dB HL.
Figures 14 through 19 are group audiograms showing averaged thresholds (both ears) for each of the participants with hearing loss. All but three listeners who participated in the study had symmetrical hearing losses.
In addition, data were collected for ten listeners without hearing loss.
Six pairs of sentences (one male and one female talker) were selected from the corpus of the Lexington Dialogue Sentences. These materials were developed at Lexington specifically for evaluation of hearing aids. The six pairs of Lexington Dialogue Sentences selected were as follows.
1. I would like to try these shoes.
What size shoes do you wear?
2. Where did you go to school?
I went to school in New York City
3. That bookcase fits in nicely with your other furniture.
I tried to find the perfect place for it.
4. Did you do anything special over the weekend?
I went to the movies and read a lot.
5. The basket is on the table.
It is filled with beautiful flowers.
6. Did you watch the movie on television last night?
No, I watched a documentary instead.
The original sentences were subjected to three types of distortion created under either live or computer simulated listening conditions. Stimuli for the reverberation plus noise condition were recorded in three separate environments: a classroom, an auditorium, and a conference room. For the teleloop noise and peak clipping conditions digital signal processing techniques were used to simulate real-world listening conditions. The following is a brief description of how the three types of stimuli were prepared:
1. Reverberation plus noise.
In three separate recording environments sentence materials were delivered from a B & K artificial mouth at successive distances from the recording microphone. The environments were selected to represent those where a person with hearing loss is likely to encounter an assistive listening system: a classroom, a conference room, and an auditorium. A high quality microphone attached to a sound level meter picked up the signals and delivered them to a digital audio tape recorder. Speech Transmission Index (STI) measurements were made immediately following the recording of each sentence pair at each microphone location. The recorded signals were then redigitized and stored onto a computer disk for presentation during the experiment.
Figure 20 shows the STI at each of the recording distances from the microphone.
Note that STI values for the conference room at equivalent distances to those for the other rooms were comparatively quite poor. This is because of the constant background noise created by the ventilation system in the conference room.
2. Induction Loop Noise
A digital recording was made of the noise created by a poorly installed induction loop. This noise was digitally mixed with the original sentences at six signal to noise ratios ranging from 0 to 30 dB in 6 dB steps. Figure 21shows the spectrum of the induction loop noise.
The noise largely consists of a background “buzz “ (see primary at 60 Hz with harmonics) and a high-frequency hiss.
Also shown in Figures 22 and 23 are the composite spectra of the male and female voices, respectively, for the Lexington Sentences that were chosen.
- Peak clipping
The signals were digitally and symmetrically clipped at six different levels down from the peak level.
The following outline summarizes the three listening conditions.
- Reverberation and background noise - expressed as distance from the recording microphone. Also shown are the corresponding STI measurements.
2.Internally Generated Induction Loop Noise - expressed as signal to noise ratio of the RMS of speech to the RMS of noise
3.Peak clipping - expressed in level down from the peak amplitude
Listeners were seated in a sound treated booth wearing TDH 49 earphones mounted in MX 41AR cushions. Stimuli were delivered binaurally at a comfortable listening level. The level of the signal was the same for both ears. Signal delivery and data collection were controlled by a personal computer.
To adjust to a comfortable listening level the listeners were instructed to indicate the level of the quiet, undistorted sentences that they felt they could listen to for a very long time.
Listeners were instructed to judge the quality of the sentences by selecting one of the following ratings.
2. Good(I would purchase a ticket for a show)
3. Marginal (I may or may not purchase a ticket for a show)
4. Unacceptable (I would not purchase a ticket for a show)
Listeners were tested in five blocks of trials for each of the listening conditions.
Sentences were always presented in female/male question/reply pairs. Pairs for each level of distortion were presented six times. The protocol consisted of a pretest phase where sentence pairs were were presented in two orders from greatest degradation to quiet and quiet to greatest degradation. Following the pretest phase, the remaining four pairs of sentences for each level of distortion were presented in random order. The final result for each listener represents an average of the four rating responses for each subcondition recorded following the pretest phase.
8.3 Results and Discussion
Figure 24 shows the range of preferred listening levels selected by the listeners who
participated in this study. The highest level of 111 dB SPL was selected by a listener with a moderate-to-severe flat hearing loss. The dB SPL values were referenced to the level of a 1000 Hz calibration tone whose RMS level is equivalent to the average RMS level of the sentences.
For purposes of statistical data management the qualitative ratings were converted to numerical values. The excellent rating was converted to a score of 4, good was converted to 3, marginal was converted to 2, and unacceptable was converted to 1. Tables 1 through 5 show the
median ratings for each of the five listening conditions. These median values were initially
divided into groups according to degree and configuration of hearing loss. It is interesting to note that the median ratings cluster together regardless of the degree and configuration of hearing loss.
Because all of the groups rated the quality of the sentences similarly, recommendations stemming from this report are independent of degree of hearing impairment. Results, however, for the 10 listeners without hearing loss were excluded in arriving at the recommendations. Furthermore, in the interest of being more inclusive, the 75th percentile was chosen over the median (50th percentile). As opposed to an average (mean or median) which does not take into account those listeners with more strict listening criteria, the ratings for 75 percent of the listeners with hearing loss were considered in arriving at the recommendations.
A minimally acceptable rating of 2.25 was chosen. This value represents the average if a listener rated at least one of the four presentations for a particular subcondition as good: marginal (2), marginal (2), marginal (2), and good (3). Tables 6 through 10 show the rating level above which 75 percent of the listeners’ scores fell for each of the five listening conditions. Values that are below the minimal criterion of 2.25 are both italicized and highlighted. For the auditorium condition the microphone distance cannot be greater than 3 feet from the talker. For the classroom condition the microphone distance cannot be greater than 2 feet from the talker. For the conference room condition all of the microphone distances were judged to be unacceptable. For the induction loop condition the signal-to-noise ratio cannot be poorer than 18 dB. Finally, for the peak clipping condition the level of peak clipping down from the peak level cannot exceed 18 dB.
Table 6. Seventy-fifth percentile ratings for the Auditorium condition. In parentheses following each subcondition are the corresponding STI values.
Table 7. Seventy-fifth percentile ratings for the Classroom condition. In parentheses following each subcondition are the corresponding STI values.
Table 8. Seventy-fifth percentile ratings for the Conference Room condition. In parentheses following the subcondition are the corresponding STI values.
4 1/2 feet
7 1/2 feet
Table 9. Seventy-fifth percentile ratings for the Induction Loop noise condition.
30 dB S/N
24 dB S/N
18 dB S/N
12 dB S/N
6 dB S/N
0 dB S/N
Table 10. Seventy-fifth percentile ratings for the Peak Clipping condition.
Figure 25 shows 75th percentile ratings as a function of the STI results from the reverberation plus noise conditions combined for the three different environments. Also shown on this graph is the best fit, first order regression line to the data. For a minimally acceptable criterion of 2.25 the STI value cannot be less than .84. Note that with our criteria (a 2.25 rating for 75% of the listeners and an STI value of .84) only a few data points can be found in the unacceptable region. Note that these two data points (x’s) are for Group 5 with the greatest degree of hearing impairment.
The STI was chosen as the criterion metric over a simple distance measure because different STI values can be recorded at the same distance depending on the level of the background noise in the environment (See Figure 20).
- The system must be capable of providing 110 dB SPL and not exceed 118 dB SPL with a dynamic range on the volume control of 50 dB.
- The speech transmission index cannot be less than .84.
- The signal-to-noise ratio for internally generated noise cannot be less than 18 dB.
- The peak clipping levels down from the peak cannot be greater than 18 dB.