DSL v5
• Objectives of the DSL multistage input/output algorithm: DSL m[i/o]
• DSL m[i/o] target generation
• The output limiting stage
• Narrowband output limiting targets
• Broadband output limiting targets
• Prescription of the WDRC compression threshold (CT)
• Hearing instrument prescriptions for multi-channel compression
• Gain prescription within the WDRC stage
• Gain prescription within the expansion stage and linear stage
• DSLm[i/o] algorithm effective compression ratio
• Modifications made in the DSLv5 algorithm for adult hearing aid wearers
• Binaural fittings
• Hearing instrument prescriptions for conductive hearing loss
• Hearing instrument venting corrections
Objectives of the DSL multistage input/output algorithm: DSL m[i/o]
1. Avoidance of loudness discomfort during hearing instrument use
2. Hearing instrument prescription that ensures audibility of important acoustic cues in conversational speech as much as possible
3. Support for hearing instrument fitting in early hearing detection and intervention programs
4. Prescription of hearing instrument compression that is appropriate for the degree and configuration of the hearing loss, but that attempt to make a wide range of speech inputs available to the listener
5. Adaptation for the different listening needs of listeners with congenital versus acquired hearing loss
6. Accommodation for the different listening requirements within quiet and noisy listening environments
DSL m[i/o] target generation
In DSL v5, we use the DSL[i/o] algorithm (DSL v4.1) as a starting point, but modify it to apply WDRC to a smaller input range. The inputs selected for the WDRC range are intended to cover some or all of the conversational speech range. Low-level inputs are less likely to be included in the compression stage as hearing levels increase. The DSL multistage input/output m[i/o] algorithm includes four stages of processing: (1) expansion; (2) linear gain; (3) compression; and (4) output limiting. These m[i/o] stages reflect conventional signal processing for amplitude control in current digital hearing instruments. The final result is a series of target input/output functions that prescribe how a multi-channel, multistage device should respond to speech inputs across vocal effort levels.
The output limiting stage
Version 5 of DSL provides three variables that facilitate definition of output limiting: (a) the user's upper limits of comfort (UCL) defined with narrowband inputs, which should not be exceeded by any aided narrowband signal; (b) targets for 90 dB SPL narrowband inputs - these targets may be slightly below the upper limits of comfort if the hearing instrument is not fully saturated by a 90 dB input; (c) the broadband output limiting thresholds (BOLT) per frequency, which defined the maximum one-third octave band levels for broadband sounds. These targets may be used for slightly different purposes, depending upon the test signals at hand, and/or the user's knowledge of the signal processing characteristics of the hearing instrument to be tested. Each of these target types will be discussed in more detail in the section below.
Narrowband output limiting targets
DSL v5 provides narrowband predictions of the listener's upper limit of comfort (ULC), which may be replaced by individually measured ULCs. In either case, the fitting goal is that aided levels of high-level pure tones, warbled pure tones or speech peaks should not exceed the ULC. The predicted ULCs are limited to a maximum of 140 dB SPL in the ear canal. Narrowband targets for 90 dB SPL inputs can be generated in DSL v5. These targets may be slightly below the upper limits of comfort if the hearing instrument is not fully saturated by a 90 dB narrowband input. When verifying fit to targets using a narrowband signal, clinicians can choose to either match the 90 dB target, or to ensure that the maximum output does not exceed the ULC.
Broadband output limiting targets
DSL v5 incorporates a variable that prescribes a limiting stage for the one-third octave band levels of speech signals. This broadband output limiting threshold (BOLT) corresponds with a hearing instrument fitting that places the peaks of speech 13 dB below the upper limit of comfort. A detailed description of the rationale for BOLT is provided in Scollie et al., 2005. Clinical verification of fit to BOLT targets may not always be possible, depending upon the test signals and analyses that are currently available. This is not likely a problem if the narrowband limiting has been fitted appropriately (see above). However, the BOLT targets may be helpful in defining initial settings of programmable hearing instruments that include limiting controls for broadband stimuli - this type of setting may occur "behind the scenes" within hearing instrument programming software.
Compression
In DSL v5 we prescribe compression processing to meet the goals of providing audibility and comfortable loudness of important speech cues, given the gain limits of hearing instruments and the limited dynamic range of the individual hearing instrument user. This differs from the loudness normalization approach in previous versions of the DSL algorithm.
Prescription of the WDRC compression threshold (CT)
The DSLm[i/o] algorithm prescribes a variable CT based on hearing levels that attempts to maintain the compression stage across as broad a speech input range as possible. The intention is to support low-level speech recognition whenever possible (Jenstad et al., 1999; 2000). For more severe-to-profound hearing losses this fitting goal is modified to use WDRC as a means for controlling loudness of high-level speech. Experimental validation of this hypothesis-driven aspect of DSL v5 is necessary. Therefore, in hearing instrument manufacturers software-based implementation of DSL v5, more ambitious goals for WDRC can be incorporated by using custom CTs if the higher gains can be achieved without feedback. Figure 1 illustrates the relationship between hearing threshold levels (dB HL), the proposed input levels (dB SPL in the sound field) and the prescribed WDRC threshold from the DSLm[i/o] algorithm.
Hearing instrument prescriptions for multi-channel compression
DSLm[i/o] target calculations can be tailored to correspond to the channel structure of multi-channel hearing instruments (Scollie et al, 2005). The one-third octave band frequencies are grouped into defined channel families using the crossover frequencies of the hearing instrument. The multistage input/output algorithm is then re-computed per channel, resulting in a single compression ratio target per channel. The gains within the compression region of the revised input/output target plots are also adjusted in an effort to preserve the frequency response for mid-level signals. A slight frequency re-shaping may occur to prevent hearing instruments with different channel structures from providing a different frequency response for mid-level speech signals. Targets at moderate input levels show very little effect of channelization, while targets for very high and very low inputs show a somewhat greater effect.
Gain prescription within the WDRC stage
To prescribe gain within the WDRC stage consideration must be given to the desired range of input levels considered appropriate for amplification; the individuals residual auditory area; and the technology to be fitted. Unlike the DSL [i/o] algorithm, the DSLm[i/o] algorithm restricts the input range over which the compressive algorithm is applied from approximately 30 dB SPL to 70 dB SPL (re: FF as a function of hearing loss. A target for 60 dB SPL speech input is calculated for all one-third octave band frequencies. The WDRC stage is then defined as the straight line with a slope that equals the compression ratio target that passes through this calculated DSLm[i/o] target. For hearing losses exceeding approximately 70 dB HL a higher CT is used by the DSLm[i/o] algorithm to derive the target for 60 dB SPL speech input. Some hearing instrument manufacturers or clinicians may choose to use a lower CT (i.e., more gain for low level inputs) than recommended if the higher gains can be achieved without feedback. The WDRC stage begins at the calculated WDRC CT and ends where it meets the broadband output limiting stage. Figure 2 provides a target input/output function derived using the DSLm[i/o] algorithm compared to the target input/output function derived using the DSL[i/o] algorithm.
Gain prescription within the expansion stage and linear stage
The DSLm[i/o] algorithm calculates a default expansion threshold (ET) that is approximately 10 dB below the level of soft speech. It is assumed that input signals below this level are likely background noise and negative or no gain is desirable. The linear stage of the DSLm[i/o] algorithm spans the input range between the ET and the WDRC CT.
DSLm[i/o] algorithm effective compression ratio
The effective compression ratio calculated within the DSLm[i/o] algorithm is intended to functionally describe the amount of long-term compression of soft to loud speech inputs encountered by the listener. It is not intended to be an electroacoustic descriptor for verification, nor is it intended to be interpreted in the way that traditional compression ratios are.
DSLm[i/o] Algorithm Considerations for Individual Fittings
Modifications made in the DSLv5 algorithm for adult hearing aid wearer
The DSL[i/o] algorithm described by Cornelisse et al., 1995, and used in the DSL Method: v4.1 attempted to define the ideal amplified output for a range of input levels. The DSL[i/o] algorithm used nonlinear scaling so that input levels corresponding to the acoustic dynamic range of the normal loudness function were mapped onto the auditory area of the loudness function associated with hearing impairment, while maintaining the normal loudness relationship per frequency (Cornelisse et al., 1995). The DSL[i/o] algorithm comprised a very broad compression phase beginning at 0 dB HL. We hypothesize that the resultant gain for low- to moderate speech input levels using this approach may contribute to higher loudness levels than preferred or necessary for adult hearing aid wearers.
The DSL multistage input/output algorithm (DSLm[i/o]) used in DSL v5, does not use a loudness normalization approach for several reasons. First, current loudness models do not account for the adult-child and developmental differences required for listening reported earlier in this article. Second, loudness normalization attempts to make all sounds audible and normally loud. It is not likely that this is an appropriate goal for low-level background noise, nor is it an attainable goal given the noise floor of most hearing instruments. In developing the DSL m[i/o] algorithm we use compression processing to meet the goals of providing audibility and comfortable loudness of important speech cues, considering the general limits of hearing instruments and the limited dynamic range of the individual hearing instrument user. As discussed above, the compression stage spans as much of the range of conversational speech across vocal effort levels as possible. As a starting place, the DSL m[i/o] input range was limited to no lower than 20 dB HL for adult listeners with acquired hearing impairment. Compared to the 0 dB HL loudness normalization strategy in DSL [i/o] this provides adults with a lower level of prescribed gain and compression ratio for the entire input-output function. As shown in Figure 3, the differences in prescriptive targets are largest for mild-to-moderate losses. A smaller correction is applied as hearing loss increases which is a desired effect because it maintains audibility of speech for more severe-to-profound hearing losses for adults and children. Further experimental evaluation of this age-related correction is required, however, it appears to be in good agreement with the adult-child differences in preferred gain reported earlier in this chapter.
Binaural fittings
It appears from the literature that binaural loudness summation for low- to mid-level signals may range between 3 to 5 dB (Scollie et al., 2005). In DSL v5 prescribed targets for speech are reduced by 3 dB across input levels for binaural fittings. Clinical application of this 3 dB reduction can be either applied or omitted on an individual basis, at the clinician's discretion. For example, some clinicians may choose to omit the binaural correction in pediatric fittings in order to provide increased audibility of speech cues.
