Electroacoustic Evaluation, Part 1 (Out of the Box Setup)

The Setup

Initially I experienced difficulty uploading programs to the Tympan unit, so I carried out electroacoustic analyses using the ‘out of the box’ setup which was programmed with 8-channel WDRC. The following parameters were the defaults in the ‘WDRC_8BandComp_wExp’ code:

{317.1666, 502.9734, 797.6319, 1264.9, 2005.9, 3181.1, 5044.7}, // cross frequencies (Hz)
{0.57, 0.57, 0.57, 0.57, 0.57, 0.57, 0.57, 0.57}, // compression ratio for low-SPL region (ie, the expander…values should be < 1_.0)
{45.0, 45.0, 33.0, 32.0, 36.0, 34.0, 36.0, 40.0}, // expansion-end kneepoint
{20.f, 20.f, 25.f, 30.f, 30.f, 30.f, 30.f, 30.f}, // compression-start gain
{1.5f, 1.5f, 1.5f, 1.5f, 1.5f, 1.5f, 1.5f, 1.5f}, // compression ratio
{50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0}, // compression-start kneepoint (input dB SPL)
{90.f, 90.f, 90.f, 90.f, 90.f, 91.f, 92.f, 93.f} // broadband output limiting threshold (comp ratio 10)

The figure below shows the filter bank for the 8-channel WDRC MATLAB code on which the Tympan code is based (e.g., Alexander and Masterson, 2015).

Throughput Delay

For reasons discussed later, throughput delay was measured with the volume setting at -10 and +10 dB. Delay was measured using a Frye Fonix 7000 hearing aid analyzer with the earphone attached to a 2-cc (HA-1) coupler with putty. The test was repeated several times to ensure consistent results. The plots below show that delay was reliably 5.7 ms for each volume setting. This value is right in line with measurements of delay obtained on premium hearing aids from all of the major manufacturers in an 8-year period (Alexander, 2016; http://www.canadianaudiologist.ca/issue/volume-3-issue-4/hearing-aid-delay-feature/).

Hearing Aid Test Box Measures (HA-1 Coupler)

Acoustic measures in a sound-isolating chamber (‘test box’) and in a real ear were obtained using a Verifit (by Audioscan, A Division of Etymonic Design Inc.) hearing aid analyzer. The Tympan unit was connected to a Sony ECM-CS10 Lapel Electret Microphone and to a pair of Klipsch S4 earphones. Per manufacturer recommendations for the test box measures, the Sony microphone was centered between the loudspeakers in the test box (+/- 45 degrees) and the Verifit reference mic was placed within 2 mm of the front of the Sony microphone. The photos below show the setup.

One of the earphones was connected to the measurement (coupler) microphone via an HA-1 (2 cc) coupler using putty as one would normally do when testing custom hearing aids that fit in the canal. As shown in the following picture, the earphone, coupler, and microphone were placed outside the test box to minimize the chance of feedback occurring when testing high gain settings.

Input/Output Curves

Input-Output curves were obtained using the “ANSI S3.22-2003 Input/Output” measurement in the Verifit. In the figure that follows, curve 1 (green) is with the volume setting at -10 dB, curve 2 (magenta) is with the volume setting at 0 dB, curve 3 (cyan) is with the volume setting at +10 dB, and curve 4 (orange) is with the volume setting at +20 dB. The red circles at 50 dB SPL input correspond to the nominal compression thresholds in the Tympan unit. The colored numbers indicate the gain for the 50 dB SPL input signal. The red number near the top of each graph corresponds to the measured maximum output dB SPL for each frequency. All values are exact values and were computed using the information in the saved data file. Compression ratio (CR) was computed by dividing 20 dB by the difference in the output dB SPLs for 70- and 50-dB SPL input levels (or 15 dB divided by the difference for the 65 and 50-dB SPL input levels for tones ≥ 1000 Hz at the +10 dB volume setting, since output limiting was present for the 70-dB input level). Compression ratios were computed only for the -10 and 0 dB volume control settings because they were the only settings where the output dB SPL was below the maximum output for a 70 dB SPL input. As expected, compression ratios were similar for the two volume settings.

When comparing the measured values to the nominal values in the software, the frequency responses of the microphone and earphones needed to be considered. Overall, the 50-dB SPL gain values are closest to the nominal channel gains with the volume control at -10 dB. A volume control setting of -5 dB likely would have generated a closer correspondence. The empirical WDRC compression ratios are less than the nominal ratios, but research has shown that this is always the case, especially as compression time constants become shorter. The measured maximum output values seem reasonable for an individual with mild to moderate hearing loss. It is difficult to know what is controlling the maximum output in the software since different values appear in different places. Considering that the underlying compression architecture is based on my MATLAB code, I know that output limiting is a 2-stage process, with output limiting in each channel, followed by broadband output limiting on the summed channels. In the code, it appears as if output limiting in each channel is set to around 90 dB SPL and broadband output limiting is set to 105 dB SPL. Given this information, the code seems to do an adequate job controlling maximum output levels. Using a faster attack time should yield slightly lower values. Of course, as mentioned earlier, the ultimate SPL that matters is the SPL at the research participant’s tympanic membrane. Given that a real-ear to coupler difference of +10 dB or more is not unusual for adults at some frequencies, caution should be used, ESPECIALLY WHEN TESTING CHILDREN due to the fact that smaller residual ear canal volumes produce higher SPLs (Boyle’s law).

Distortion Tests (HA-1 Coupler)

Harmonic distortion was measured using the Verifit “Distortion” test. First, 65-dB SPL tones were tested with the volume knob set at -10, 0, 10, and 20 dB. Then, the same tests were performed using 80-dB SPL tones. From the figures below (note the values on the ordinate auto scale), it is clear that minimal distortion is evident except at high output levels (e.g., high input levels and/or high volume control settings). Given the threshold settings for output limiting compression, I suspect that the harmonic distortion is due to desired behavior of the software rather than the physical limits of the earphones. This concern will be addressed later in a test condition that applied gain, but not compression.

##ANSI S3.22-2003 (HA-1 Coupler)
Standard hearing aid tests of quality control using the ANSI S3.22 (2003) standard were performed with the volume setting at -10 and +10 dB since the input/output curves indicated that a +20 dB setting would likely put the Tympan unit in output limiting compression and the distortion measures indicated that significant harmonic distortion would result. No attempts were made to put the hearing aid at its loudest, linear, or reference testing settings per ANSI test requirements. The purpose of these tests was to evaluate the overall frequency response, gain, maximum output (OSPL 90), and compression time constants at typical user settings.

Keeping in mind that relative to gain for channels ≥ 800 Hz that gain for channels ≤ 500 Hz was 10 dB less and that gain for channels between 500-800 Hz was 5 dB less, the frequency response for the 60-dB test tones (magenta line) at the -10 dB volume setting is fairly flat with peaks that seem to correspond to 5 of the 7 crossover frequencies in the 8-channel compressor (317, [503], 798, [1265], 2006, 3181, and 5045 Hz). Considering that FIR filters are used to create the channels and that they add in phase, +6 dB peaks are expected at the crossover frequencies when narrowband or tonal signals are used. The peaks can be even greater when the tones near these frequencies are processed by 2 independent compressors (as in the present case) because each compressor ‘sees’ a lower signal level when the frequency is on the slope of its filter and therefore applies gain appropriate for this level. The figure below shows the same phenomenon in the MATLAB version of the code (e.g., Alexander and Masterson, 2015). In fact, the frequency response of the OSPL90 curve (90-dB test tones, green line) takes on a similar shape as that for the 60-dB tones, which more than likely reflects the multichannel output limiting compression. As discussed later, tests were conducted on a condition that did not have channelization or compression suggests that the peaks might actually be acoustic resonances.

The average gain for the 50- and 60-dB tones and the output levels for the 90-dB tones (see inset) are roughly consistent with the information shown by previously described input/output curves. Interestingly, the reported attack and release times at 1000 Hz and above are exactly those that were set in the code. Finally, the equivalent input noise (28 dB SPL) is not atypical of levels that my Hearing Aids I classes from the last 8 years have measured in hearing aids with premium technology. It should also be taken into consideration that equivalent input noise depends on the overall frequency response of the device with respect to the test frequencies and that the earphones+coupler microphone setup was outside of the sound-isolating chamber.

For the +10 dB volume setting, the results are more erratic, especially for the 60-dB tones since the volume setting was 12 dB in excess of the ANSI recommended level (known as, ‘Reference Test Gain’). However, this volume setting more closely represents the recommended level (known as ‘Full On Gain’) when measuring the OSPL90 curve. Comparing the OSPL90 values and curve to those for the -10 dB volume setting, it is apparent that most of the increase in maximum output can be attributed to the lower frequencies. In addition, the average gain at 50-dB (44 dB) is consistent with the information obtained from the input/output curves. Therefore, it seems that when the mic+Tympan+earphone combination is not driven into saturation, that the frequency response is mostly flat and that when driven into saturation, it shows a low-pass characteristic, which I believe is not uncommon with actual hearing aids. The low-pass characteristic at high output levels is also consistent with the findings of Audette (2017a, http://openaudio.blogspot.com/2017/05/calibrating-my-earphones-with-tympan.html). This will be discussed in more detail later. Finally, while the equivalent input noise level (36 dB SPL) is greater than that for the -10 dB volume setting, it is similar to some of the super-power hearing aids that my classes have tested in the past. In addition, as noted before, the earphones+coupler microphone setup was outside of the sound-isolating chamber.

##Equivalent Input Noise (EIN)
To look specifically at the effect of the volume setting and equivalent input noise, the ANSI test was re-run at each volume setting between -25 and +20 dB. This time, the earphone and coupler microphone were inside the test box, except for the +20 dB volume setting because of audible feedback. Three test runs were conducted at each volume setting. For each run, only the EIN was documented. The mean EIN across the three runs are shown in the table below. In the reference section of this report is an article by Holder et al. (2016), which indicates that (a) the amount EIN with the Tympan unit is not much different from modern hearing aids, and that (b) this measurement is prone to error due to the relative lack of sound isolation provided by the Verifit test box. The data also indicate that EIN is unaffected by volume settings between -10 and +10 dB and that it increases by a few dB as the volume settings are increased or decreased beyond this range. If the dominant noise source were in the microphone, which is the most common source of internal noise in hearing aids, one would expect EIN to increase and decrease with the volume setting. The fact that EIN is mostly unaffected by the volume setting suggests that the dominant source of the internal noise is after the volume control (e.g., the amplifier or DAC).

##Pink Noise Mulitcurve (HA-1 and HA-2 Couplers)
Broadband noise was used to evaluate output levels and gain across frequency as function of input level. Pink noise was used because it has a falling frequency characteristic similar to speech (-3 dB/octave vs. -6 dB/octave) and because it has a flat frequency response when analyzed with 1/3-octave filters. Responses were measured with the volume setting at -10 and +10 dB, although I forgot to save the data for the latter with the HA-1 coupler. Therefore, data will be presented for the -10 dB volume setting using the HA-1 coupler and for both volume settings using an HA-2 coupler, which I was experimenting with at the time.

For the HA-1 coupler and -10 dB volume setting, the frequency response to a broadband input seems to be relatively flat between at least 300-4000 Hz, excluding the various peaks, which I believe are associated with channel summation and/or acoustic resonances. This conclusion is also based on the facts that gain was reduced ≤ 800 Hz and that the frequency response of the coupler significantly rolls off above 4000 Hz. It is likely that the negative gain at the very lowest frequencies is due a ‘slit leak’ between the earphone tip and the coupler. The fact that these same frequencies also have a reduced response with the HA-2 coupler (discussed later; but a little less than with the HA-1 coupler) suggests that acoustic leakage is NOT solely responsible for this finding. It can be seen in the figures below that the overall frequency response does not change much as input level is increased and that gain systematically decreases with increasing input level, which is expected since wide dynamic range compression was active.

The picture to below shows the HA-2 coupler that is designed to connect to the earhook of a behind-the-ear hearing aid. The long tube is supposed to represent the tubing between the earhook and the terminal end of the earmold. I experimented with this coupler by removing the earbud from the earphone and connecting it directly to the coupler. Since I failed to save the data for the +10 dB volume setting with the HA-1 coupler, I will report the raw data obtained with the HA-2 coupler and then report HA-1 equivalent data that was computed by taking the average difference between the output levels obtained with the HA-1 and HA-2 couplers for the -10 dB volume setting.

The most obvious difference between the HA-1 and HA-2 couplers at the -10 dB volume setting is the resonant peak at 4000 Hz associated with the HA-2 coupler. As mentioned earlier, the slightly better low-frequency response that was obtained with the HA-2 coupler might reflect the tighter connection between the earphone and the coupler. Otherwise, the same trends are shown in the data as the input level changes. Looking at the data for the +10 -dB volume setting, it is clearly noticeable that output is being limited as the input level is increased. This effect occurs for high frequencies before the low frequencies, thereby giving the overall frequency response a low-pass characteristic as the system is driven into saturation.

To get a fuller description of the change in frequency response across output levels, the HA-2 data for the two volume settings were combined into one plot and transformed into HA-1 equivalent responses as described above. The following figures show the transformed data using dotted lines for the different input levels when the volume setting was -10 dB and solid lines when the volume setting was +10 dB. The figure showing the SPLs suggests that the change in frequency response is due to the overall output levels and not to the volume setting per se since there is a nice overlap in the data between the two volume settings (i.e., the 85-dB input level for the -10 dB volume setting an.d the 55-dB input level for the +10 volume setting). In case there was any question, the 85-dB response with the -10 dB volume setting does not show any limitation in output, therefore, the change in the frequency response is not due to the microphone. Similarly, because the gain is greatest for the lowest input levels, the amplifier in the Tympan unit is not a limiting factor.

Until I was able to implement and test software changes, it was unclear whether the limited output levels was due to the earphones or the software settings for broadband output limiting threshold (BOLT), which was set to about 90 dB SPL across channels. Given that the BOLTs were slightly higher for channels ≥ 2000 Hz, suggests that ability to achieve greater output levels at lower frequencies might simply be a physical property the earphones (i.e., more mass than stiffness). Alternatively, it could be the case that output limiting was activated for the high frequencies before the low frequencies due to the rising frequency response of the Sony microphone (http://openaudio.blogspot.com/2017/05/calibrating-microphones-with-tympan.html). This hypothesis is based on the fact that the distortion data presented earlier for high output levels indicated that substantial distortion was present only in the low frequencies. Therefore, it could be that output limiting kept the high frequencies below the physical saturation limit of the earphones, but not for the low frequencies, resulting in an increase in low-frequency output levels accompanied by distortion.

The first figure below shows the HA-1 equivalent gains across frequency for the two volume settings. The figure that follows shows the difference between the two gain settings for each input level. For the 55-dB input level, the difference is exactly as advertised, about 20 dB across frequency, except in the low frequencies where ambient noise may have interfered with the measures. As expected from the previous measures, progressively smaller differences in high-frequency gain are realized as input levels increase.

##Speechmap (HA-1 Coupler), Body Aid
To evaluate how the Tympan unit will amplify a real speech signal, a ‘speechmapping’ was conducted using the Verifit and an HA-1 coupler. The standard test signal in the Verifit is a 12.8-second calibrated speech signal with the same long-term average spectrum as the ‘international average’ (Byrne et al., 1994). The measured signal applies several correction factors to simulate a real hearing aid fitting. The first correction factor is the microphone location effect, which accounts for how the frequency response of the test signal at the microphone inlet is shaped by its location on the ear, in the ear, or on the body. A ‘body aid’ was chosen as the hearing aid type in the Verift because a lapel microphone is currently being used with the Tympan unit. As shown by the figure below (Audioscan, 2016), the body aid has the least frequency shaping of the styles. This differs from the frequency response of the Sony microphone measured by Audette (2017b, http://openaudio.blogspot.com/2017/05/calibrating-microphones-with-tympan.html), which shows a rising frequency response with about a 20 dB difference between the low and the high frequencies.

The other correction factor is the real-ear to coupler difference (RECD), which accounts for the difference between the how the ear and the coupler shape the sounds. The idea is to estimate the SPL of the amplified sound at the level of the tympanic membrane. There are several factors that influence this measure, which are beyond the current discussion. The residual ear canal volume (between the sound outlet and the tympanic membrane), which roughly correlates with the age of the individual, is a major determinant. Shown below are age-corrected averages for the RECD (based on tabular data in Audioscan, 2016). The average for adults was automatically applied by the Verifit to the SPL measured from the coupler microphone. As shown in the figure, this adds about 12 dB to the measured response between 4000 and 6000 Hz. Note that individuals, especially children, differ along this measure so extreme caution must be used when presenting high-level input signals to an individual if there is any uncertainty about the maximum output levels of the device. To help avoid presenting an overly intense signal to a research participant, simple code could be written that would allow the Tympan unit to present pink noise for the purposes of obtaining an RECD. The RECD could then be used to first test the unit in the test box.

Speechmaps were obtained with -10 and +10 dB volume settings for soft (55 dB SPL), average (65 dB SPL), and loud (75 dB SPL) input levels. Maximum Power Output (MPO) was also measured using 85-dB SPL tones. The changes in output levels between the -10 and +10 dB volume settings are similar to the differences observed for pink noise. Namely, the differences in output for the soft speech are about 20 dB, while the differences are smaller for the MPO test signal, especially in the high frequencies. Interestingly, while the differences for the loud speech are also smaller, they are smallest for the mid frequencies instead of the highest frequencies.

##Fitting Range
From my experience, with some adjustments to the frequency shaping, the dynamic range shown above for the Speechmap obtained with the -10 dB volume setting would easily accommodate most mild-to-moderate hearing losses. An easy way to show this is to enter hypothetical hearing losses into the Verifit and it to compute and overlay the prescriptive target levels for each of the tests. For demonstration purposes, flat hearing losses were entered into Verifit and Speechmaps obtained for volume settings ranging from -10 to +10 dB. Thresholds were 55 dB HL for the -10 dB setting and increased 5 dB for each 5 dB increase in volume. In the figures that follow, the solid red line corresponds to hearing thresholds converted into dB SPL at the tympanic membrane for an average adult ear. The solid lines (green, magenta, and cyan) correspond to the measured output of the Tympan unit expressed in dB SPL at the tympanic membrane for soft (55 dB SPL), average (65 dB SPL), and loud (75 dB SPL) presentation levels. The color-coded plus (+) signs indicate the corresponding prescriptive fitting targets for each input level according to the Desired Sensation Level (DSL) v5.0a method for adults. The solid orange line corresponds to the measured maximum output levels for 90-dB pure tones. The orange symbols correspond to the 90-dB target levels while the black asterisks correspond to the estimated loudness discomfort levels. As shown by the figures that follow, with a little frequency shaping and adjustments to output limiting compression thresholds, hearing losses up to 70-75 dB HL can easily be accommodated by the Tympan unit and Klipsch earphones. Combined with previous data, it seems that it can amplify soft speech with relatively modest internal noise and high-level narrowband sounds with relatively low distortion for these losses.

##Pink Noise and Speechmap (Real Ear), Body Aid
Pink noise responses and Speechmaps using a probe microphone (see pictures) were obtained in the ear of a normal-hearing person for only the -10 dB volume setting and only for soft, average, and loud speech. The higher volume setting and MPO were not tested in order avoid the loudest discomfort. It should be noted that the Verifit has a feature that will automatically stop a test if a certain dB SPL is detected in the ear canal. In the upper left-hand corner of the plots below, it can be seen that the safety level was set at 115 dB SPL. In addition, notice that a body aid was selected since the lapel microphone was clipped to the middle of the individual’s shirt. While, the real ear responses for both pink noise and speech do not reveal anything too surprising, they do demonstrate that the Tympan unit can be tested like a regular hearing aid.