Pure-tone sweeps (or sinusoids) have been used as the stimulus to verify the performance of hearing aids tested with couplers and worn on real-ears for many years. In recent years, however, more acoustic stimuli have become available as possible alternatives to pure-tones for verifying the performance of hearing aids. While some may expect similar outcome from the use of different stimuli in verification, the reality is that the choice of a particular stimulus can lead to outcomes that may alter ones interpretation of the result of verification. This article illustrates some of the effects of using different signals to verify todays hearing aids.
Acoustic Stimuli Used in Verification
There are many reasons for verifying the performance of a hearing aid. Some dispensing professionals verify in order to ensure that the chosen device matches the manufacturers specifications. This is frequently done using standardized hearing aid couplers in order to ensure the integrity of the hearing aid and its components. Other dispensing professionals verify in order to ensure that the real-ear gain/ output of the device matches the desired recommendations of a prescriptive target. This task implicitly assumes that a certain level of real-world satisfaction may be expected when the target output/gain is met. To achieve the above objectives, three types of acoustic stimuli that differ in their spectral and temporal complexity have been used in the hearing aid clinics:
Pure-tones: Pure-tones are simple signals with energy at only one frequency. Pure-tone sweeps (or sinusoids) are pure-tones with continuously changing frequencies presented at the same fixed level over the range of audible frequencies in a fixed amount of time. Pure-tones and pure-tone sweeps are simple because their frequency and intensity levels can be precisely controlled by the test equipment and no special frequency analyzers are needed for their measurement. The signals have been used to estimate the frequency-gain/output characteristics of hearing aids. In addition, they have also been used to determine the input-output characteristics, saturation distortion, attack/release times, as well as the current drain of hearing aids. If the purpose for verification is to determine the integrity of the structural components in a hearing aid, this simple stimulus can serve its purpose quite well. Indeed, the most recent ANSI standard (ANSI S3.22-1996)1 recommends the use of sinusoids in testing hearing aid performance for quality assurance.
On the other hand, since only one frequency is presented at a time, any interaction among frequencies that are likely to occur in nonlinear signal processing cannot be examined using sinusoids. Furthermore, since the validity of verification is limited to the specific stimulus that is used, one cannot generalize the performance of nonlinear hearing aids to complex sounds using results obtained from sinusoids. Thus, pure-tones are not suitable if the purpose of verification is to examine the performance of the hearing aids to real-life complex signals.
Speech: Speech is a complex broadband signal that has varying spectral and intensity content over time. On the surface, speech is a good stimulus for testing hearing aids because it is the signal that most hearing aid wearers would want to hear in their daily environments (ie, high face validity). Because of its complex nature, any interactions among frequencies may be revealed by this stimulus.
The problem with using real speech as a test signal is that speech cannot be easily standardized; its spectrum varies substantially across talkers, gender, age, etc. Furthermore, the frequency and intensity content of real speech cannot be controlled easily. Such unpredictability and variation in temporal and spectral content over time makes it an unreliable and imprecise verification tool for the purposes of quality control and target matching. For the same reason, real-life sounds should not used to verify performance of hearing aids or for target matching purposes (although they may be acceptable for fine-tuning hearing aids).
Composite Signals: In order to test a hearing aid at the frequencies of speech without the noted variation in frequency and intensity content, artificial signals that have the same long-term spectral characteristics of speech are synthesized. These are called speech spectrum shaped composite signals. The advantage of using composite signals is that many frequencies can be tested in a short time, and that interactions among frequencies may be tested. In addition, they can be synthesized with controlled characteristics so that specific processing algorithms on a hearing aid may be optimally evaluated. This advantage makes composite signals the most appropriate signal for evaluating todays (and future) nonlinear hearing aids. With careful interpretation, one can use the results obtained from composite signals to predict the performance of the hearing aids to real-life complex signals.
There are many different speech spectrum shaped composite signalseach synthesized with different criteria so they are similar, but not identical, to real speech. In one commercial speech spectrum shaped composite signal modeled after the ANSI S3.422 recommendation, the composite speech is synthesized to have frequency components from 100 Hz-8000 Hz spaced in 100 Hz intervals. The amplitudes of these individual frequency components are controlled to achieve similar long-term characteristics as speech recorded directly in front of the talker. This has a slope of -6 dB/octave.
Another speech spectrum shaped composite signal is the ICRA signal (International Collegium of Rehabilitative Audiology) that purports to approximate the ideal long-term far-field speech spectrum which has a slope of -9 dB/octave. A reason for using the ICRA signal is that many of the recordings are modulated at rates typical to that of conversational speech. This may be useful when examining nonlinear hearing aids with noise reduction algorithms.
Another example is a dynamic stimulus where sweeping sinusoids over the speech range are modulated to resemble the dynamic range of typical speech. This type of composite signal may be useful for analyzing the dynamic response of a hearing aid to speech-like signals.
There are many more speech spectrum shaped composite signals available commercially, each with its own unique design and spectral/temporal characteristics. Because the purpose of this paper is to raise awareness of the potential problems with stimulus differences and not to provide an exhaustive demonstration of the different stimuli, we will focus on the responses of hearing aids to only three signals: 1) the pure-tone sweep (Tone); 2) the continuous speech spectrum shaped composite signal according to ANSI (ANSI), and 3) the modulated speech spectrum shaped composite signal according to ICRA (ICRA).
Figure 1. Frequency spectra of the stimuli (measured in a 100 Hz bandwidth) used in the demonstration; pure-tone sweep at 50 dB SPL (green), unmodulated ANSI signal (blue), and modulated ICRA signal (black).
Figure 1 shows the pure-tone sweep at 50 dB SPL and the spectra of the ANSI and ICRA signals at an overall level of 70 dB SPL that are used in all the following demonstrations. These signals are generated from the Frye 6500 Hearing Aid Test system, and all subsequent output/gain measurements are performed with a default bandwidth of 100 Hz. Clearly, the ICRA signal has more low-frequency energy and less high frequency energy than the ANSI signal.
Figure 2a-B. Top (2a): Gain curves of a linear hearing aid (Logo) tested with a pure-tone sweep (at 50 dB SPL), a composite-speech noise (70 dB SPL overall), and the ICRA noise (70 dB SPL). Bottom (2b) : Output curves of the same linear hearing aid tested with the same signals.
Response of Linear Hearing Aids to Different Stimuli
At any particular frequency, a linear hearing aid provides the same gain at all input levels for that frequency. A direct consequence is that the measured gain of a linear hearing aid should be identical regardless of the stimuli used for the measurement. Figure 2a shows the frequency-gain curves of a linear hearing aid to the ANSI signal at an overall level of 70 dB SPL, the ICRA signal at the same overall level, and a pure-tone sweep at 50 dB SPL. Note that the frequency-gain curves for all three signals are identical despite the differences in their input spectra.
In other words, gain on a linear hearing aid is insensitive to stimulus level difference (since gain is independent of input level). A corollary is that any stimuli may be used to measure gain of a linear hearing aid. This unique property of linear hearing aids may have helped the use of linear prescriptive formulae where most specify target gain based on the degree of hearing loss.
While gain of a linear hearing aid is insensitive to stimulus level difference, output from a linear hearing aid is sensitive to level differences among stimuli. This is logical since output is the sum of the input level and gain. Figure 2b shows the output of the linear hearing aid to all three stimuli. It is fair to say that the linear hearing aid yields different output for the different stimuli despite the same gain setting.
The use of stimuli with different spectral characteristics could be problematic during verification when one tries to match the output of the hearing aids (linear and nonlinear) to a prescriptive target. This is because the recommended output level at a certain frequency offered by a prescriptive formula is not only determined by the degree of hearing loss, but also on the intended input signal and on the bandwidth of the measuring device (or hearing aid) within which the level is specified.
Ideally, one should use the same stimulus during verification as that used in the target formulation in order to legitimize the match. The use of a different signal would violate this apples-to-apples assumption and make the appropriateness of the matched output questionable. This is demonstrated in Figure 2b where the same hearing aid setting yields three different frequency output curves depending on the stimuli used. (Authors note: To be most accurate, fitting targets recommending a desired output (viz. not gain) should also recommend the bandwidth of the measurement system (eg, 1 Hz, 100 Hz, 1/3 octave, etc) that one can use to verify the output of the hearing aids. This is because different measuring bandwidths would yield different frequency-output curves with a complex signal).
This discrepancy had not raised concern until recently because many linear and nonlinear prescriptive formulae specify gain targets rather than output targets; sinusoids rather than complex signals are used as the stimuli for verification in these cases. On the other hand, if we consider that human ears respond to output rather than to gain, it is logical to include prescriptive targets based on the output of hearing aids. Furthermore, a standardized signal (eg, ANSI or ICRA signal) and a standardized measurement bandwidth (eg, 1/3 octave) must be specified to minimize variability in outcome. This will become more important when nonlinear hearing aids are concerned.
Thankfully, some of these concerns are beginning to be widely addressed today.
Nonlinear Hearing Aid Responses
Effect of multi-channel nonlinearity: The advent of nonlinear hearing aids brings new considerations to the choice of acoustic stimulus for verification. A nonlinear hearing aid provides different gain at different input levels. Typically, more gain is provided to low-input levels, and less gain is provided as the input level increases (except when the hearing aid is in nonlinear expansion).
Figure 3a-b. Top (3a): Frequency gain curves of a 15-channel hearing aid (Diva) tested with a pure-tone sweep (at 50 dB SPL spectral level), a speech-composite noise (70 dB SPL overall), and the ICRA noise (70 dB SPL). Bottom (3b): Frequency-output curves of the same hearing aid tested under identical conditions. The noise reduction algorithm on the aid has been deactivated.
The effect of nonlinear gain is especially seen in a multi-channel device which has independent gain control at each frequency channel. It suggests the possibility that more variation in the frequency-gain (and frequency-output) curves may be possible when different stimuli are used to test the hearing aid. This is shown in Figure 3a and 3b where a 15-channel nonlinear hearing aid is tested with the same stimuli used previously (a pure-tone sweep, the ANSI signal, and the ICRA signal). Figure 3a shows that more high-frequency gain is reported with the ICRA signal than the ANSI signal, possibly because the ICRA signal has less high frequency energy than the ANSI signal (thus more gain because of the lower input). This is in contrast to the linear hearing aid where all three stimuli show the same frequency-gain curve. As expected, the frequency-output curves (Figure 3b) show significant output differences among the three stimuli used to determine the output responses.
Effect of special signal processing algorithms: Many nonlinear digital hearing aids also employ special noise reduction algorithms that seek to identify the nature of the incoming signals to make additional gain adjustment beyond that provided by compression. For example, the Senso Diva hearing aid analyzes the level distribution of the input signal to make such a determination in each of its 15 frequency channels.3 Analysis that shows a bi- or multi-modal level distribution would suggest speech in that channel. A uni-modal distribution, which indicates that the signal intensity is relatively constant over time, would suggest a noise (or non-speech) input. When a speech signal is identified, the assigned gain for the specific input level will be maintained. However, when noise is identified along with speech, the gain of the hearing aid will be reduced from the assigned value depending on the estimated signal-to-noise ratio and the speech importance weighting of the specific frequency. With this hearing instrument, it takes about 10 seconds for the algorithm to confirm the identity of the incoming signal.
This kind of speech/noise analysis has two notable effects on verification (for both coupler and real-ear measures). Because the continuous ANSI signal lacks the intensity fluctuation over time that is typical of real speech, it would be interpreted as noise by the hearing aid, resulting in gain reduction.
Figure 4a-b. Top (4a): Frequency-gain curves of a 15-channel hearing aid tested with a pure-tone sweep (at 50 dB SPL spectral level), a speech-composite noise (70 dB SPL overall), and the ICRA noise (70 dB SPL). Bottom (4b): Frequency-output curves of the same hearing aid tested under identical conditions. The noise reduction algorithm on the aid is active and the signals have been presented for 15 seconds.
Figure 4 shows how this hearing aid responds to the pure-tone sweep, the continuous ANSI signal, and the modulated ICRA signal when each has been presented for 15 seconds. In principle, the modulated ICRA signal has similar temporal fluctuation as speech and will be identified by the aid as speech. In contrast, the continuous ANSI signal lacks the modulation and will be identified as noise by the noise reduction algorithm. This is seen in Figure 4a where the gain curve obtained with the ANSI signal is 10-12 dB lower than that obtained with the ICRA signal. It is of interest to note that the gain curve obtained with the pure-tone sweep which does not reflect the real-life performance of the hearing aidis significantly different from the other two gain curves.
On the other hand, when examining the output curves in Figure 4b, one sees the combined effect of modulation and spectral level differences between the ICRA and the ANSI signals. The output curve obtained with the ICRA signal is much higher than the ANSI output curve in the low and mid frequencies but only slightly higher than the ANSI output curve in the high frequencies. This, again, points to the difference between examining gain and output responses.
Another variable that may affect the output of a digital hearing aid with a noise-reduction algorithm is the duration of the stimulus used. Recall that the noise-reduction algorithm needs a certain time window to identify the nature of the input. It takes additional time to fully realize its effects. This means that a noise signal that is presented for less than the duration of the time window may not demonstrate (or only partially demonstrate) the full efficacy of the noise-reduction system. That is, modern day hearing aids are sensitive to the duration of the input stimulus as well. This is the case of the unmodulated ANSI signal, and it will not be true with the modulated ICRA signal, or when the ANSI signal is modulated.
Problems with Target Matching
Careful choice of stimulus is also important if one tries to match the gain/output of hearing aids to generic prescriptive gain target. Kuk & Ludvigsen4 indicated that, because of channel summation, the real-life output (ie, relative to complex signals) of multi-channel hearing aids matched to a prescriptive target using pure-tone sweeps will be higher than the real-life output of single channel hearing aids matched to the same target.
Figure 5a-b. Top (5a): Frequency-output curves of 4 hearing aids 15 channel Diva, 3-channel Senso C+, 2-channel bravo, and single channel Logomatched to a NAL-R target using pure-tone sweep as stimulus; Bottom (5b): Frequency-output curves of the same 4 hearing aids when the ANSI speech spectrum shaped composite signal was used (at 60 dB SPL overall) as the stimulus.
In general, the more channels there are and the higher compression ratio (CR) in each channel, the more noticeable this summation effect. This is seen in Figure 5a where four hearing aidsthe 15-channel Senso Diva, the 3-channel Senso C+, the 2-channel Bravo, and the single-channel Logoare matched in output to a NAL-R target using a pure-tone sweep presented at 60 dB SPL. Figure 5b shows the output of the same four hearing aids to the ANSI signal that has an overall level of 60 dB SPL (noise reduction deactivated). The output curve of the 15-channel Senso Diva is almost 10 dB higher than that of the single-channel Logo, and over 5 dB higher than those of the 2-3 channel aids. This means that even though all four hearing aids meet the gain target for pure-tones, they yield different output levels (and possibly loudness) when presented with complex signals.
In general, the higher the number of channels, the louder it will sound. This suggests that unless one makes provision in the prescriptive gain targets for channel summation, one should perform target matching with a complex or composite signal, and match for output in order to minimize the channel summation effect to real-life sounds.
Other Practical Implications
If different stimuli (and durations of the same stimulus) result in different outputs from a nonlinear hearing aid, one cannot but ask which is the optimal stimulus for use to verify the output of hearing aids for the purpose of target matching. Using the optimal stimulus is necessary so that one will not reject or mis-adjust an otherwise satisfactory hearing aid on the grounds that it does not match target specifications. A corollary is that one may match a gain target in the clinic only to find the hearing aids to be too loud (or too soft) in real-life.
Another practical problem may be the choice of different stimuli across clinics. A clinical site that uses a particular stimulus may find a chosen hearing aid to match a specific target whereas another site, using a different stimulus, may find the hearing aid to miss the same gain target. The second site may either return or mis-adjust the hearing aid. This is a possibility especially when there are so many hearing aid (and/or real-ear measurement) systemseach with a different measurement bandwidth and each having many different stimuli and targets for the clinicians to choose. Such variability would be confusing for anyone, and it may even compromise the fitting.
So, What Should We Do?
It is obvious that the choice of stimulus has a significant impact on results when performing hearing aid verification. To decrease variability and to increase accuracy, it is important that we use the same optimal stimulus for verification. This means that one should use the same stimulus on which the prescriptive target is based, and measure the output with a measurement system that uses the same bandwidth as recommended by the prescriptive formula. This would ensure that the hearing aid is optimized for the specific listening condition used during verification. Hopefully, the results would generalize to other real-life situations. The following lessons learned should be considered when choosing a stimulus if the aforementioned conditions cannot be realized.
Judicious use of pure-tone signals: While pure-tone signals have been used for many years for purposes such as quality assurance, evaluating attack/release times, and determining input-output curves, they do not reveal the processing of the hearing aid to complex signals. Furthermore, it may introduce artifacts (eg, blooming in the low-frequency) and compromise the goodness of target matching to generic formulae. On the other hand, if the hearing aid user complains that high-frequency sounds are too loud, using a high frequency sinusoid may be better than using a speech-shaped composite signal to verify the output of the hearing aid. But, in general, caution should be taken when using pure-tones if the goal is to estimate the performance of the hearing aid to real-life sounds.
Use of composite/complex signals: Because complex signals allow one to examine the interaction among frequencies (including intermodulation distortion, channel summation, etc) and they are the sounds that one hears in everyday environments, they should be used for testing todays nonlinear hearing aids. Use of this type of signal should be acceptable for linear hearing aids as well.
On the other hand, one should not use complex signals such as real-speech for target matching (real-ear or coupler) because of the difficulty in controlling the stimulus characteristics. For target-matching purposes, use a well-defined composite signal recommended by the specific prescriptive target.
Similarly, it would be extremely helpful if the developers of any fitting formula (generic or device-specific) specify the characteristics of the speech spectrum they used to formulate their targets, as well as the measurement bandwidth (1 Hz, 1/3 octave, etc) that one should use to measure the output of hearing aids. Manufacturers of hearing aid test systems (including real-ear) should also indicate how the output of hearing aids is analyzed. This would minimize variations in the measured output and increase the accuracy of hearing aid fitting.
Agree on a composite signal: As illustrated before, there are many types of composite signals, each with its unique temporal-spectral characteristics. For example, the ANSI speech signal has more high-frequency energy than the ICRA signal; conversely, the ICRA signal has more low-frequency energy than the ANSI signal. Different stimuli result in different frequency-gain/output curves. This can hinder communication and lead to mis-interpretation across clinics if each uses a different stimulus. It is important for the profession to agree on a standard signal for the purpose of verification and target matching. Currently, the ANSI S3.32 (1992)2 composite speech spectrum shaped signal has been the recommended standard. Other stimulus such as the ICRA signal which is shaped according to the idealized speech spectrum specified in ANSI S3.5 (1997)5 has been proposed as an alternative.6
At the very least, clinicians need to specify the stimulus that they used for verification and check with the manufacturer of the device on the appropriateness of their chosen stimulus. If the stimulus may not be the most appropriate and if the clinician is not in a position to acquire new test stimuli, then he/she may at least be warned of the discrepancy and take this into consideration when interpreting the test results.
Examine both the output and the gain: As illustrated before, examining the gain of a hearing aid could be deceiving. Because the ear hears the output of a hearing aid and not its gain, it is important to consider the output of the hearing aid during verification as well. For example, it would be more meaningful to examine if the outputmeasured in a similar manner as the ear would analyze sounds (eg, critical band or 1/3 octave)is within the wearers residual dynamic range and not whether it meets a particular gain target.
Know your hearing aids and the purpose of verification: As indicated earlier, some digital nonlinear hearing aids have noise reduction algorithm that may give a faulty impression of their performance when tested using conventional stimuli like pure-tones or unmodulated ANSI signal. Especially with these hearing aids, one should fully understand the purpose for verification. If the purpose is to examine the frequency response of the hearing aid in quiet, one may deactivate the noise reduction algorithm during verification, present the stimulus for only a brief period, or use a stimulus that will not activate the noise reduction algorithm. If one is interested in examining the effect of the noise reduction algorithm, then the algorithm should remain activated, and a non-modulated composite noise should be presented for at least 10 s-20 s before one examines the noise-reduction effect.
The availability of different acoustic stimuli is driven by the types of signal processing available in todays hearing aids. By its very nature, advances in technology always lead progress in standardization. Thus, the observations we have made may only be valid today and be outdated tomorrow. In this ever-changing industry, one thing that is certain is that the next generation of DSP hearing aids will challenge the appropriateness of the stimuli we have just discussed.
The authors thank Denise Keenan and John Nelson for assisting with the illustrations in this article.
|Francis Kuk, PhD, is director of audiology at the Widex Office of Research in Clinical Amplification-USA, Lisle, Ill, and Carl Ludvigsen, MS, is manager of audiological research at Widex A/S, Vaerloese, Denmark.|
Correspondence can be addressed to HR or Francis Kuk, PhD, Widex Hearing Aid Co, 35-53 24th St, Long Island City, NY 11106-4116; email: [email protected].
1. American National Standards Institute. Specification of Hearing Aid Characteristics (ANSI S3-22 1996). New York: Acoustical Society of America;1996.
2. American National Standards Institute. Testing Hearing Aids with a Broad-Band Noise Signal. (ANSI S3-42 1992). New York: Acoustical Society of America;1992.
3. Kuk F, Ludvigsen C, Paludan-Muller C. Improving hearing aid performance in noise: challenges and strategies. Hear Jour. 2002;55(4):34-46.
4. Kuk F, Ludvigsen C. Variables affecting the use of prescriptive formulae to fit modern nonlinear hearing aids. J Am Acad Audiol.1999; 10(8):458-465.
5. American National Standards Institute. American National Standard Methods for the Calculation of the Speech Intelligibility Index. (ANSI S3.5- 1997). New York: Acoustical Society of America;1997.
6. Dreschler W, Verschuure H, Ludvigsen C, Westermann S. ICRA noises: Artificial noise signals with speech-like spectral and temporal properties for hearing instrument assessment. International Collegium for Rehabilitative Audiology. Audiology. 2001; 40(3):148-157.