12 Lessons Learned About Linear Frequency Transposition

AE has been used extensively for almost 3 years. Here’s what we have learned.

Frequency transposition has been proposed for almost 40 years¹; yet, as with ideas like directional microphones and open-fit instruments, it has been delayed widespread implementation within hearing aid platforms. Following a large research and development effort, Widex reintroduced this form of signal processing in its Inteo ISP hearing aid as the Audibility Extender (AE). With almost 3 years of clinical and laboratory experiences, we have seen that, when the right candidates are chosen and when they are fit following specific guidelines, the AE algorithm provides a viable solution for adults and children with an unaidable hearing loss in the high frequencies.

FIGURE 1. SoundTracker screen showing that the frequency region to be transposed (yellow shaded area) does not receive amplification.

This article summarizes our experience, and we hope the information will spark further interest in the proper use and evaluation of such an algorithm.

Rationale for Frequency Lowering

Frequency lowering means making the information carried by the high frequency fibers available to the lower frequency fibers. For example, information carried by fibers responsible for 4000 Hz may now be “lowered” to fibers at 2000 Hz for decoding. There are two major forms of frequency lowering techniques—frequency compression in which the whole frequency spectrum or a portion of the frequency spectrum is compressed to fit into a narrower frequency region; and frequency transposition in which the higher frequency is moved down in frequency to a lower frequency region. For a more detailed description of the rationale, please see Kuk et al.²

12 Facts and Findings About Linear Frequency Transposition

1 Implementation with AE and the critical nature of the “start frequency.” The AE used in the Inteo hearing aid uses linear frequency transposition. This means sounds above a particular frequency (which is deemed unaidable because of the non-responsiveness of the high frequency inner hair cells, or unreachable because of insufficient hearing aid output) are moved down in frequency to a lower frequency without compressing it into a narrower range.

The AE is the only frequency lowering algorithm that uses transposition instead of compression. In addition, its action is unconditional, meaning that the algorithm is activated continuously. Some frequency compression algorithms either compress the entire speech spectrum when voicing is not detected, or compress the mid- and high-frequency information on an unconditional basis to fit into the mid frequency region.

Frequency transposition means no amplification in the transposed area. The rationale for transposition (and frequency lowering in general) is that the high frequency region is “dead” where no amount of amplification will restore hearing. Thus, in designing the AE algorithm, the input signals that are aidable and unaidable are treated differently. The dividing line for that treatment is called the “start frequency” of the AE algorithm. Frequencies above the start frequency are to be transposed lower. They are assumed to be unaidable; thus, no amplification (or gain) is provided. On the other hand, because the frequency region below the start frequency is aidable, amplification is provided so it may be amplified to an audible level.

The situation is best illustrated in Figure 1 with the Sound Tracker³ used in our fitting software. As a quick reminder, the height of the bars represents the output of the hearing aid from each one of the 15 channels. Output that exceeds the sensogram (blue solid line) represents the output of the hearing aid in that channel that is audible. The yellow-shaded area is the region to be transposed. In Figure 1, 2500 Hz is the start frequency. One can see that below 2500 Hz, gain is provided to each of the channels (darker bars) to result in the final output above the sensogram. However, no gain (darker color bars) is visible above 2500 Hz. Instead, the colored bars are now moved to the lower frequency channels (1200, 1600, 2000 Hz) where the transposed high frequencies would be.

It is important to recognize that transposition means no amplification is applied in that specific region. If we have a start frequency in an aidable region, that frequency region (and higher) will not be amplified but instead be replaced by transposed sounds.

Subjectively, the wearer may find the AE program to sound unnatural, especially during the initial fit. In addition, the wearer may perform worse with the AE program than with the conventional program (or master program) because the AE has replaced a totally functional region with an artificial stimulus. This is why the proper choice of start frequency is critical in the success of the AE. A detailed discussion of the start frequency is available from Kuk et al.⁴

FIGURE 2a-b. Individual audiograms for adult (2a, left) and pediatric subjects (2b, right).

2 Frequency transposition is the same thing as calculated distortion (constructive distortion); it should not be used if there are better alternatives. A reason why too low a start frequency or too much transposition (or compression) is not desirable is because frequency lowering is a form of calculated distortion. This means that the input signal is altered purposely to provide the hearing-impaired persons with the additional cues for processing.

An analogy may be made to cochlear implants. The stimuli or electric “clicks” from a cochlear implant are dramatically different from the natural sounds and are meaningless to the typical normal-hearing person; yet, with sufficient training and motivation, these “clicks” convey sound information to the cochlear implant wearers. This suggests two major implications:

• One should use transposition only when there are no better alternatives. If a frequency region can be amplified by conventional means, one should amplify that region instead of using transposition to access the same information.
• When using AE, one should use the least amount of transposition (or the highest start frequency) to achieve the objective. This will help preserve the natural cues within the original speech signal and minimize the amount of distortion.

3 AE is efficacious in adults and children. We conducted several studies to examine the efficacy of the AE algorithm in both adult and pediatric subjects. With adult subjects, we have also collected data on hearing-impaired subjects and normal-hearing subjects with a simulated hearing loss. The results supported the efficacy of frequency transposition:

FIGURE 3. Percentage of adult subjects preferring the AE (over the master or no AE) for bird songs, music, and female continuous discourse during the final visit.

• Korhonen and Kuk⁵ used 10 normal-hearing subjects and evaluated their identification of voiceless consonant nonsense syllables filtered above 1600 Hz, with and without frequency transposition. After each test trial, subjects were trained for 15 minutes on the transposed sounds with which they had the most difficulty. The use of normal-hearing subjects minimizes any biases from the hearing loss configurations, and cognitive or plasticity issues of the subjects.
• Kuk et al⁶ have also completed data collection on adult subjects with a high frequency hearing loss (Figure 2a) who wore binaural Inteo IN-9 hearing aids in an open-ear, thin-tube mode. The rationale was that this group of patients would be more likely to utilize the transposed information. These subjects were tested at different times and were provided different types of training.
• Auriemmo et al⁷ studied the efficacy of the AE on children attending the Special School District of St Louis County, Mo. This group of subjects had more hearing loss than the adult subjects (Figure 2b). The protocol for testing these children was similar to that of the adults. Children were trained on the use of the AE, as well as participated in aural and speech rehabilitation. For both adults and children, identification of nonsense phonemes on the Nonsense Syllable Test (NST) was utilized as well as non-speech stimuli such as bird songs and music.

4 Differential effectiveness de­pending on stimulus complexities. We included in our evaluations several types of stimuli that varied in spectral complexity. They included stimuli like bird songs, which were simple in spectral content. We also included music passages and continuous discourse passages, which were spectrally more complex. These stimuli were presented in a paired-comparison format for the subjects to indicate their preference (ie, with or without transposition).

The results of our evaluation varied depending on the type of stimuli. Figure 3 shows the percentage of adult subjects selecting the AE for each of the three stimuli during the final visit. The preference for AE was highest when bird songs were used as the stimulus (over 90%), followed by music stimuli (70%), and speech stimuli (60%).

The preference for bird songs is easy to understand. Without the AE, bird songs would not be audible because of the extent of the hearing loss. With the AE, what was previously inaudible became audible at a lower frequency. With music and speech stimuli, the interaction between what was previously audible (without transposition) and the transposed sounds could lead to an “unnatural” perception for some wearers.

The implication to the differential effects of AE is that the efficacy of AE should be evaluated with speech and non-speech stimuli. A battery of stimulus materials, including everyday sounds and nature sounds, would be appropriate for a comprehensive evaluation.

The inclusion of everyday sounds and nature’s sounds in the evaluation of frequency transposition becomes more relevant as the wearer’s hearing loss becomes more severe to profound. This is because of the increased reliance that these patients placed on hearing environmental sounds for safety and comfort, even though the ability to hear those sounds may not reflect their ability to hear conversational speech.

FIGURE 4a-b. Improvement in consonant identification score on the NST at a soft (30 dB HL) and average (50 dB HL) input level for adults (4a, top) and children (4b, bottom).

5 Improvement in consonant identification. An expected desirable outcome of frequency transposition is to provide an improvement in speech intelligibility over the best available conventional amplification. That was the outcome of the studies we conducted on the AE in adult and pediatric wearers. Kuk et al⁶ evaluated the efficacy of the AE on 13 hearing-impaired adults with a sloping and precipitously sloping high frequency hearing loss. After 1 month’s use of the AE, the average subject improved consonant identification by 10-15% over the master program at a soft input level (30 dB HL) and by 5-10% at a conversational level (Figure 4a).

Auriemmo et al⁷ reported the efficacy of the AE on 10 hearing-impaired children ages 6 to 12 years. These children were trained on the use of the master program for 3 weeks and then underwent 6 weeks of exclusive use of the AE program. After 6 weeks of AE use, consonant identification of the average child improved by 20% at a 30 dB HL input level (from 48% to 68%) and 5-10% at a 50 dB HL input level.

FIGURE 5. Accuracy of /s/ and /z/ production on a reading and a conversation task using the child’s own hearing aids, the Inteo master program, and the Inteo AE program.

6 AE is most effective in improving voiceless consonant sounds. An advantage of using a nonsense syllable test and scoring on a phoneme basis is the ability to analyze the types of errors that the wearers make. This helps understand how AE processing affects speech understanding. For adult subjects with a simulated hearing loss, voiceless consonants such as the /s/, /sh/, /ch/, /th/, /p/, and /t/ were consistently better identified with transposition than without frequency transposition.⁵ Obviously, these are sounds with significant high frequency content. Considering that the subjects used in these studies had primarily a high frequency hearing loss, it is expected that most of the improvements would be for the high frequency phonemes. The same observations were also evident in hearing-impaired adult and pediatric subjects.

7 Vowels are not negatively affected. One likely outcome of frequency transposition is the masking of lower frequency sounds, such as vowels. Fortunately, this has not been observed in either the adult or pediatric subjects. In the adult study, the final average vowel score was 88% with the master program and 92% when the AE program was used at a 50 dB HL presentation level. In the pediatric study, the final vowel scores were 95% with the master program and 100% with the AE program.

The fact that vowels are not affected negatively is fairly predictable: the start frequency for transposition in the majority of subjects was higher than 2500 Hz. If one considers that (in English language) the frequency ratio of the first two formants determines the identity of the vowel and the second formant of the vowel with the highest frequency (eg, /i/) is lower than 2500 Hz, then transposition does not alter the formant ratio. On the other hand, for a more severe loss in the low frequencies or for a low start frequency, it is possible that vowels are affected initially. This may be a result of the alteration between formant relationships, or from potential masking effect of the vowels by the transposed high frequency sounds.

An implication for the lack of effect suggests that vowels may not be used in the evaluation of AE efficacy when the hearing loss is primarily in the high frequencies. Their use may be justified when a more severe loss in the low frequencies is encountered.

FIGURE 6. Frequency of use (% choosing AE) and individual hearing loss at 4000 Hz of the adult subjects. A moderate correlation was observed between hearing loss and frequency of use.

8 Ensures audibility of soft sounds. One of the findings in both the pediatric and adult AE studies is that the magnitude of AE improvement was greater for the 30 dB HL level than for the 50 dB HL level. For example, in the adult data, an AE benefit of 10-15% was seen at the 30 dB HL input level but only 5-10% at the higher 50 dB HL input level. This suggests that the AE has the potential to ensure better audibility of soft sounds.

Previously we showed in Figure 4a the adult consonant scores obtained with the master program and the AE program measured at 30 dB HL and 50 dB HL input levels. At the 50 dB HL input level, the master program reached a 62% correct score, whereas a score of 68% was reached with the AE program (6% improvement). At the 30 dB input level, a score of 50% was reached with the master program and a score of 65% was reached with the AE program. This shows that the score (around 65%) measured with the AE program at a 30 dB HL input level was similar to that measured at a 50 dB HL input level; whereas the same was not true with the master program (50% at the 30 dB HL vs 62% at the 50 dB HL input). This suggests that the AE may be another way to ensure audibility of soft sounds.

9 Improvements in speech production in children. While there is plenty of evidence in the area of cochlear implants to suggest that cochlear implantation leads to better speech perception and improved speech production in children, this area of evidence is seldom explored in hearing aid use.

In the pediatric study,⁷ we examined the consistency and accuracy of the child’s production of the /s/ and /z/ sounds during a 5-minute, open-ended, storytelling task and reading task. The child’s speech production was recorded and later analyzed by a speech pathologist skilled in phonetic transcription. The number of times the child produced the /s/ and /z/ phonemes correctly was counted and compared to the potential number of times he or she could have produced these sounds. Figure 5 shows the accuracy of the production at the baseline (with own aid), with the master program at 3 weeks of use, and with the AE program after 6 weeks. The average accuracy score was around 70% with the children’s own hearing aids (which were all digital hearing aids). Performance improved to around 80% with the Inteo master hearing aid after 3 weeks’ use. Performance further improved to almost 90% with the use of the AE program. This was statistically significant (P < 0.05).

The improvement in speech production skills further highlights the importance of good audibility in speech and language development in children. It also indicates the need to include such measures when evaluating the efficacy of hearing aids with this type of processing.

FIGURE 7. Fitting range of the Inteo Audibility Extender algorithm.

10 The more severe the hearing loss, the greater the acceptance and benefit. In both the adult and pediatric studies, we tried to examine if the degree of hearing loss correlates with the use and benefit of the transposition program. In the adult study, we allowed subjects to select their preferred program (either master or AE) during daily use. The frequency of use of a particular program was logged by the Sound Diary datalogging mechanism within the Inteo.⁸

Figure 6 shows the frequency of use of the AE program as a function of the hearing loss measured at 4000 Hz. Adult subjects with more hearing loss at 4000 Hz used the AE program more frequently than those with less hearing loss at 4000 Hz. A significant correlation was noted. Children with more hearing loss in the low-to-mid frequencies reported greater benefit with transposition.

These observations from the adult and pediatric subjects lead one to tentatively conclude that the benefit of transposition (as measured by improvement in speech identification and/or frequency of use) may be directly affected by the degree of hearing loss. For the most effective use of frequency transposition, the frequency region for transposition must not be aidable by conventional amplification. This usually means a severe-to-profound degree of hearing loss in the high frequencies above 3000 to 4000 Hz.

Indeed, the more they cannot be helped by conventional means, the more likely the AE algorithm provides unmatched benefit. Another important aspect is the frequency region to where the sounds are transposed (ie, below the start frequency). In principle, this region should be aidable with no more than a moderate-to-severe degree of loss. Although the pediatric data suggest that more speech intelligibility improvement can be expected with a more severe loss, there are a few caveats to remember:

• The degree of resolution (frequency and temporal) becomes poorer as the degree of hearing loss increases. Consequently, the ability of the remaining auditory fibers to utilize the transposed signal becomes questionable as the hearing loss below the start frequency becomes more severe.
• The transposed signal is amplified to a level where it can be audible. The more hearing loss at the target frequency, the more gain will be needed to amplify the transposed signal to audibility. If the chosen hearing aid has limited gain at the target frequency (or if it has a low MPO at the target frequency), the transposed sounds may not be usable.

11 Audiometric candidates for the AE. Thus, in using the audiogram as a criterion to select AE candidates, one should make sure that there is a dead (or unaidable) region—the region from which information needs to be transposed. Typically, the hearing loss would be greater than 70 dB HL at and above the start frequency. Below the start frequency, the hearing loss must be aidable but not too severe. This could range from normal hearing to a moderately severe degree in the low-to-mid frequencies. A 70 dB HL limit may be used as a criterion. Figure 7 shows the verified fitting range of the Inteo AE algorithm.

12 AE benefits improve over time. The effect of frequency transposition on subjective acceptance and speech understanding may not be immediately recognized by all wearers at the initial fitting. Such benefits take time to “unfold.”

Indeed, no hearing aid feature has demonstrated as strong an acclimatization effect as frequency transposition. Figure 8 demonstrates the consonant and vowel identification scores of the pediatric subjects over time, showing performance with their own aids (at the initial visit), the Inteo master (initial and after 3 weeks), and the AE program (initial, 3 weeks, and 6 weeks).

FIGURE 8. Acclimation to AE: Vowel and consonant identification scores over time at a 30 dB HL presentation level. The “own aid” and “master baseline” data were measured at the initial visit, and the “master post AT” and “AE baseline” were measured at the same time, 3 weeks after use of the master program. The “AE post AT1” was measured 3 weeks after use of the AE, and the “AE post AT2” was measured 6 weeks after use of the AE.

In the consonant scores, a sharp increase in performance is seen when subjects switch from their own aids (18%) to the Inteo hearing aid (48%, master baseline). This demonstrates the superiority of the Inteo processing over the children’s own digital hearing aids. The improvement from the master to the AE program, however, is more gradual and increases at each visit, suggesting that subjects acclimatize to the benefit of the AE gradually. Such acclimatization effect is not evident when comparing the “master baseline” and the “master post AT,” which was collected 3 weeks after use of the master program.

Another observation can be made with vowel identification. As expected, the initial improvement in vowel scores reflects the superiority of the Inteo processing over the subjects’ own hearing aids. What is interesting to note is the decrease in vowel scores at the initial AE use (AE baseline, from 92% to 88%), only to return to almost 100% when the AE is used for 3 weeks. This reflects possible confusion by the wearers when sounds are initially transposed. It also suggests that extended use of the AE can overcome the initial confusion and lead to additional improvement in speech identification.

Such observations of improvement over time have several implications. First, the benefit of the AE takes time to be fully realized. Second, it suggests the possibility that a wearer’s initial subjective impression of the AE may not be entirely informative or predictive of what the wearer may experience after a period of AE use. This means that one must not make unnecessary changes in the fitting until the wearer has a chance to experience the sound processing of the hearing aid. The question is how can one be sure that the recommended settings are the best to convince the wearers to follow our recommendation?

Verification is the only way to ensure optimal settings. Briefly, the Sound Tracker may be used to ensure that the transposed sound is above the in situ threshold (sensogram) of the wearer when one uses high frequency stimuli (such as birdsongs, high frequency speech sounds, including /s/ and /sh/). If audibility is ensured despite some initial objections, one should proceed with the default settings without adjustment.

Linear frequency transposition: extending the audibility of high frequency information, by Kuk F, et al. September 2006 The Hearing Review.

Proper counseling and training are also important to fully realize the benefit of the AE. However, the format of the training may not be as critical as long as the wearers are motivated to explore the potential of the AE feature in hearing everyday sounds. Kuk et al⁹ provided a discussion on how to facilitate the acceptance of the AE while ensuring an optimal fit.

Conclusions

This paper summarizes some of the important lessons we have learned on the Inteo frequency transposition since it was introduced almost 3 years ago. Although specific to the Widex Inteo product, some of the same lessons may also apply to devices using other frequency lowering techniques. Suffice to say, when the right candidate is selected and the AE fitted properly, linear frequency transposition can further enhance the hearing of people with an unaidable/unreachable high frequency hearing loss.

References

1. Braida L. Hearing Aids: A Review of Past Research on Linear Amplification, Amplitude Compression, and Frequency Lowering. ASHA. 1979; Vol 19 [monograph].
2. Kuk F, Korhonen P, Peeters H, Keenan D, Jessen A, Andersen H. Linear frequency transposition: extending the audibility of high frequency information. Hearing Review. 2006;13(10): 42-48.
3. Kuk F, Damsgaard A, Bulow M, Ludvigsen C. Using digital hearing aids to visualize real-life effects of signal processing. Hear Jour. 2004;57(4): 40-49.
4. Kuk F, Keenan D, Peeters H, Korhonen P, Hau O, Andersen H. Critical factors in ensuring efficacy of frequency transposition I: individualizing the start frequency. Hearing Review. 2007;14(3): 60-67.
5. Korhonen P, Kuk F. Use of linear frequency transposition in simulated hearing loss. J Am Acad Audiol. 2008. In press.
6. Kuk F, Peeters H, Keenan D, Lau C. Use of frequency transposition in thin-tube, open-ear fittings. Hear Jour. 2007;60(4):59-63.
7. Auriemmo J, Thiele N, Marshall S, Pikora M, Quick D, Stenger P. Effect of frequency transposition in school-aged children. Poster presented at: American Academy of Audiology annual convention; April 2008; Charlotte, NC.
8. Kuk F, Bulow M. Short-term data-logging: another approach for fine-tuning hearing aids. Hearing Review. 2007;14(1):46-53.
9. Kuk F, Keenan D, Peeters H, Lau C, Crose B. Critical factors in ensuring efficacy of frequency transposition II: facilitating initial adjustment. Hearing Review. 2007;14(4):90-96.