Objective: Hearing aids are more and more technically advanced, but do not necessarily guarantee the reproduction of useful signals in all working environments. This is particularly the case for speech intelligibility. This study focuses on the prediction of hearing aid performance in the case of a moderate deafness setting, in service and industrial work environments. To improve intelligibility, hearing aids propose signal processing options such as noise reduction and compression. These processes can transform hearing aids into nonlinear systems. The aim of this study is to develop a nonlinear method for the characterization of hearing aids. Materials and Methods: The method is based on the synchronized swept sine (SSS) signal method. The SSS method is applied for determining hearing aid frequency responses fitted according to the present methodology and several processing options. The characterization of hearing aid’s program containing the noise reduction function is specifically analyzed. Indeed, to be fully active and efficient, the hearing aid, with the noise reduction feature activated, needs to be immersed in a noisy environment which does not allow nonlinear characterization. A linear approach is taken to study this feature. Three hearing aids commonly sold by hearing care professionals are studied here; all of them have three different programs. The characterization for each program is discussed. Results: The statistical study showed that the intelligibility, assessed using the speech transmission index in these sound environments, is well estimated for every program, although certain differences are observed when the compression effect is too high in the service work sector. Conclusion: The characterizations of hearing aids using the programs studied did not highlight the presence of frequency nonlinearities. The characterization method could not take into account amplitude nonlinearities when there is too much gain compression in the hearing process. Globally, all the hearing aid programs provided a very significant improvement in intelligibility in service and industrial work contexts.
Keywords: characterization, hearing aids, intelligibility, noisy working environment, synchronized swept sine
|How to cite this article:|
Malrin A, Ducourneau J, Chevret P. Characterization and Prediction of Speech Intelligibility at the Output of Hearing Aids in a Noisy Working Environment. Noise Health 2023;25:183-94
|How to cite this URL:|
Malrin A, Ducourneau J, Chevret P. Characterization and Prediction of Speech Intelligibility at the Output of Hearing Aids in a Noisy Working Environment. Noise Health [serial online] 2023 [cited 2023 Nov 30];25:183-94. Available from: https://www.noiseandhealth.org/text.asp?2023/25/118/183/386590
| Introduction|| |
According to the World Health Organisation, 1.5 billion people are affected by auditory deficiency and 430 million of them require rehabilitation services. From now to 2050, it is estimated that 700 million people will need such rehabilitation. Different types of deafness can stem from genetic causes, infectious diseases, aging, or occur following exposure to high sound levels. In the work place, deafness can reduce the efficiency to carry out tasks and have impacts on safety as it hinders the perception of sound signals, the understanding of speech in the presence of noise, and the localization of sounds.
Therefore, in the context of work in industry (noisy environment with machines) or in the service sector (office environment), workers affected by deafness may be in danger and be subject to permanent stress due to their disability. In order to maintain an accurate perception of the sound environment and allow workers with impaired hearing to carry out their work in complete safety, efficiently and independently, a possible solution is for them to wear hearing aids in their workplace. Indeed, hearing aids can contain a large number of functionalities designed to improve the speech intelligibility and auditive comfort of persons with impaired hearing, such as noise reduction, speech enhancement in noisy environments, microphone directivity, and the anti-Larsen effect. These treatments are increasingly sophisticated and technical data relating to the operation of hearing aids are strictly protected by their manufacturers. It is therefore impossible to know the nature of the processes they use, especially as to whether these processing options are linear or not.
Moreover, hearing aids are not programmed to be used specifically in a noisy working environment. Some functionalities, such as noise reduction, can even deteriorate speech intelligibility despite reducing listening effort. According to a study by the IRSST, workers wearing hearing aids question the benefits regarding intelligibility, audibility, and hearing protection, and whether to wear their hearing aids at work or not. Furthermore in, hearing health professionals do not all agree on the issue and criticize the lack of information necessary to respond to the question of the benefits of hearing aids at work.
The objective of this study is therefore to propose a characterization method that takes into account the presence of possible frequency nonlinearities in the operation of hearing aids and to provide some answers concerning the real benefits provided by hearing aids at work, and specifically to the question of speech intelligibility.
This paper presents two experimental protocols that were specifically developed in the framework of this study. The first is used to determine the nonlinear frequency responses of hearing aids when the noise reduction feature is turned off. In the case where the noise reduction option is active, linear approach is considered because the hearing aid must be immersed in a noisy environment to be well characterized. Indeed, the presence of noise during the characterization process does not allow the identification of nonlinearities with the synchronized swept sine (SSS) characterization method as shown in the section 2.2. In the following, these methods will be referred as hearing aid characterization with noise reduction turned on and hearing aid characterization with noise reduction turned off.
The second protocol is aimed at simulating the operation of hearing aids based on the two characterizations performed (with noise reduction turned on or turned off). In such a case, it is referred as simulation. The aim is to simulate the hearing aid output signals in noisy working contexts (sector and industrial environments). The simulation conditions will be described later in the paper in the section 2.3.3.
The quality of the simulations can be estimated by making comparisons with measurements made at the output of the hearing aid.
This paper is divided into two main parts. In the first one, the characterization is explained. The experimental setup and the sound stimuli are also presented. In the second part, the simulation stage is explained. Comparisons are then made between the simulations and measurements performed at the output of the hearing aids. Moreover, the possible benefit provided by hearing aids regarding speech intelligibility in noise is discussed on the basis of comparisons of speech transmission index (STI) values between the input and output of hearing aids.
| Characterization stage|| |
A hearing aid is a complex system that incorporates signal processing algorithms that can often be very different from one brand to another. The analysis of this type of system thus requires the implementation of adapted experimental methods making it possible to study the behavior of any hearing aid whatever its brand.
To estimate speech intelligibility in the workplace for a hearing-impaired employee wearing hearing aids, it is necessary to characterize the hearing aid, for example, via its frequency response. In addition, there does not appear to be any method in the literature for characterizing hearing aids taking into account the possible presence of frequency nonlinearities. There are several methods that allow identifying the response of nonlinear systems. For instance, there are methods that use random or pseudo-random signals,, and methods that use series of sinusoidal functions.,, In this paper, the SSS method initially introduced by Farina, then extended by Novak, is applied. The principle of this method will be described in the following.
As mentioned in the introduction, hearing aids that use noise reduction algorithms must be characterized when the feature is fully active. Although the feature can be activated even if there is no noise around, it needs to be fully activated during the characterization process. In 2004, Hagerman and Olofsson proposed a method to separate noise from useful signals at the output of the hearing aid. However, the use of this method assumes that the system is linear and is therefore not applicable on a nonlinear system. Thus, to study this specific option, a linear study of the hearing aid is proposed by applying the phase-inversion method of Hagerman and Olofsson with the SSS used in this case as a useful signal.
In this paper, the quality of the simulation phase is performed exclusively through speech intelligibility. To do this, it is possible to make use of psychoacoustic index such as the Short Time Objective Intelligibility (STOI) measure, the Hearing-Aid Speech Perception Index (HASPI), the Speech Intelligibility Index SII, etc. In this study, the STI was chosen. The STI was chosen because it is a fully validated index and it can take into account the quality of the room through the reverberation time. The study presented here does not currently take into account the impact of reverberation on intelligibility. Research work taking this effect into account is currently under development and will be presented later.
Hearing aid characterization with noise reduction turned off
The SSS developed by Novak permits the nonlinear characterization of a system by solving the polynomial Hammerstein model. It is important to mention that this method allows the identification of frequency nonlinearities only. Nonlinearities in amplitude are not detected by this method. [Figure 1] illustrates the SSS method.
|Figure 1 Block diagram of the characterization of a nonlinear system using the synchronised swept sine method (up) and a representation of the polynomial Hammerstein model (down),|
Click here to view
In this figure, the SSS signal s(t) is a sinusoidal signal whose frequency increases exponentially as a function of time. It is used to excite the nonlinear system under study. On one side, an inverse filter signal
is built. On the other side, signal y(t) is measured at the output of the system. Lastly, the convolution operation, called nonlinear convolution in, between signals
and y(t) permits calculating a set of characteristic impulse responses (denoted hn(t) with n≥1) of the nonlinear behavior of the system under study. The Fourier transform of the impulse responses hn permits calculating the filters Hn(f) representing the nonlinear behavior of the system in the frequency domain, with filter H1 corresponding to the linear response of the system, filter H2 corresponding to the response of the first harmonic, etc. The filters Hn(f) are then modified using a matrix transform, in order to solve the polynomial Hammerstein model. An important property of the polynomial model of Hammerstein is that all filters Hn(f) are linear. Therefore, this model can’t take into account amplitude nonlinearities but frequency nonlinearities only.
This experimental protocol is used to describe the hearing aid when the noise reduction feature is turned off. Therefore, this method will be applied to the characterization of programs P1 and P3 which are described later in the paper.
Hearing aid characterization with noise reduction turned on
One of the processing options implemented very frequently in hearing aids is the noise reduction functionality. Its purpose is to improve the signal-to-noise ratio (SNR) at the hearing aid output in the antrum auris of the user, so that speech intelligibility increases and that the noise masking the sound is reduced. The hearing aid processes the signals it receives by reducing the surrounding noise. Thus, it identifies the “useful” signal (like the speech signal) and the “nuisance” signal (masking noise, cocktail party, or machine noise). To characterize a hearing aid with the noise reduction functionality activated, it is necessary to immerse the hearing aid in a noisy environment and simultaneously emit the SSS signal.
Hagerman and Olofsson developed a technique that allows separating the useful signal from the noise at the output of a hearing aid in order to study the noise reduction process. The method is based on the successive emission of the superimposition of a useful signal (the SSS signal) and a noise signal (noted V) and then the useful signal and noise in phase inversion. Nevertheless, this technique assumes that the system being tested is linear which is not compatible for the determination of the higher orders filters Hn (n>1) solution of the Hammerstein polynomial model.
Indeed, let’s examine this by considering two signals at the input of a Hammerstein model one after the other:
The outputs of the model are the respective signals:
Where * designates the convolution operation.
These expressions can be written using the Newton’s binomial formula and become:
If we add or subtract y1(t) and y2(t), we obtain:
So, by applying the sum or the difference between y1(t) and y2(t), all the coupling terms cannot be eliminated for n>1. To determine properly the filters Hn only, the signal SSS has to be used without noise interactions. Therefore, the filter Hn(n>1) associated with the SSS signal cannot be determined and it is not possible to use the SSS method to analyze the frequency nonlinearities of the hearing aid using noise reduction function.
In contrast, if the hearing aid, when detecting noise, applies strictly different treatments to the noise and speech signals, then the separation method of Hagerman and Olofsson could be used together with the SSS method. Since it is difficult to know in advance whether the hearing aid processes noise independently of the useful signal, a linear approach is taken by combining the Novak and Hagerman methods, but only filter H1 will be considered which correspond to the linear response of the system. If the linear characterization does not provide the correct estimation of signals at the output of the hearing aid, then it will be possible to conclude that certain nonlinear processes occur when the hearing aid uses the noise reduction feature. On the other hand, if the signals are estimated correctly at the hearing aid output, thanks to the linear characterization, it is possible to conclude that the behavior of the hearing aid using noise reduction is linear.
A block diagram of the characterization stage for the noise reduction feature is illustrated in [Figure 2].
|Figure 2 Block diagram summarizing the characterization of the hearing aid for the noise reduction feature (V = noise)|
Click here to view
are built and emitted separately at the input of the hearing aid. They correspond respectively to the sum of the SSS signal and a noise signal on the one hand and the subtraction of SSS(t) and V(t) (which corresponds to the sum of the SSS(t) signal and the inverted noise) on the other.
are recorded at the output of the hearing aid. We obtain:
with signals SSS′(t) and V′(t) corresponding respectively to SSS(t) and V(t) distorted by the hearing aid. e1(t) and e2(t) represent error signals such as internal noise, distortion, or interactions between the SSS signal and V. Lastly, by calculating the sum and difference of signals
, it is possible to isolate signals SSS′(t) and V′(t).
On the basis of these two equations, provided that errors e1(t) and e2(t) are sufficiently weak, the useful signals SSS′(t) and masking signals V′(t) at the output of the hearing aid can be isolated efficiently. It was specified in that if the signal distortion ratio is lower than −20 dB, then the noise due to error signals e1(t) and e2(t) is negligible. This signal distortion ratio was monitored during the characterization stage to ensure that the separated signals were estimated correctly.
Finally, signal SSS′(t) is used to determine the H1 filter with the Novak method and a noise transfer function is obtained by calculating the ratio between the Fourier transforms of the two signals V′(t) and V(t).
Experimental setup and sound equipment
Three brands of RIC (Receiver In the Canal) type hearing aids available commercially, denoted HA-A, HA-B, and HA-C were studied. These hearing aids were from three different manufacturers. They were adjusted for moderate hearing loss of the 1st degree (average tone loss of 43.7 dB HL). They each offered three different programs denoted P1, P2, and P3, all in omnidirectional mode. The [Figure 3] shows the hearing loss values as a function of frequency.
|Figure 3 Hearing loss values implemented in the hearing aids as a function of frequency|
Click here to view
(1) P1: gain from the NAL NL 2 method, no MPO (Maximal Power Output) nor noise reduction,
(2) P2: gain from the NAL NL 2 method, no MPO, with noise reduction,
(3) P3: gain from the NAL NL 2 method, compression, MPO, no noise reduction.
The NAL NL 2 (National Acoustic Laboratories − NonLinear) gain prescription method was used to adjust the hearing aid. This method allows taking into account the recruitment phenomenon, the age, gender, and experience of the hearing-impaired employee with hearing aids.
The hearing aids were placed on a KEMAR manikin (Knowles Electronics Manikin for Acoustic Research) equipped with symmetrical ear pinnae (anthropometric pinnae: GRAS KB5001 and KB002) to study two different hearing aids at the same time. Using a KEMAR manikin allowed taking into account directly the head shadow effect, the directional effect of the pinna, the interaural differences in time and intensity, and the spectral indexes. In the hollow of each artificial ear, a GRAS 26AS ¼” measurement microphone simulated the tympanic membrane and captured the sounds via an IEC 60318-4 type coupler simulating the auditory canal. The hearing aids studied were equipped with an offset earphone inserted in a closed silicon customized earmold that occludes the auditory canal. Thus, the microphones located in the manikin recorded only the signals emitted by the earphone of the hearing aid.
The manikin was installed in a semianechoic chamber at the INRS acoustics laboratory at the center of five loudspeakers. [Figure 4] shows an example of a hearing aid and the experimental setup. The dimensions of the semianechoic chamber are 11 × 7 × 7 m, its cut-off frequency is 88 Hz.
|Figure 4 Experimental setup with the KEMAR manikin at the center of 5 loudspeakers|
Click here to view
Thirty-centimeter thick wedges made of polyester wool were placed on the floor to obtain conditions close to free-field. The loudspeaker placed in front of the manikin was used to emit “useful” signals such as the SSS and speech signals. The four other loudspeakers were correlated (they emit the same signal at the same time) and used to emit masking signals, so as to immerse the manikin in a uniformly noisy environment. A ½” reference microphone was placed above the head of the KEMAR in order to control the levels of all signals emitted (SSS, speech, and masking noise). Two microphones were placed above each ear of the KEMAR, very close to the microphone system of each hearing aid (see [Figure 4], lower-right photo), to record the input signals of the device. These two microphones also ensured that the manikin was placed symmetrically in the center of the multidiffusion system. To do this, an SNR was evaluated at each ear and the orientation of the head adjusted until the difference between the two ratios was lower than 0.5 dB. This SNR was calculated by the emission with the international speech test signal (ISTS) followed by white noise. Moreover, the attenuation provided by the silicone earmold which is inserted in the auditory canal has to be checked. Hearing aid turned off, a white noise was emitted by the loudspeakers and the attenuation between the input and the output of the auditory canal was determined. If the measured attenuation was less than 10 dB over the studied frequency domain, the earmold was then repositioned and the attenuation test repeated.
- Signals used for the characterization of P1 and P3 programs:
- For the service sector environment, signal SSS was emitted at a level of 57 dB(A), which approximately corresponds to a normal speech level at 1 m from a speaker according to the ANSI 3.5-1997 standard,
- For the industrial work environment, signal SSS was emitted at a level of 69 dB(A), which approximately corresponds to a loud speech level at 1 m from a speaker according to the ANSI 3.5-1997 standard.
- Signals used for the characterization of P2 program (with noise reduction turned on):
In order to take into account the nature of the noises sources in the two environments studied, two different kinds of noise were chosen and combined with the signal SSS to characterize the hearing aid.
- For the service sector environment, the noise added to the SSS signal was a white noise whose spectrum was modified to obtain one close to that of the long-term speech spectrum (LTASS) using the ISTS spectrum. To characterize the hearing aid in this type of environment, the signals (SSS + noise (LTASS)) were all emitted at a level of 57 dB(A),
- For the industrial work environment, the noise signal was a pink noise. Indeed, in a workshop environment, the noises to which workers are subject are generally of low frequency. To characterize a hearing aid in this type of environment, the signals (SSS + noise (pink)) were all emitted at a level of 69 dB(A).
An SNR at the input of the hearing aid of 0 dB(A) between the SSS signal and the noise was chosen for the characterization stage, whatever the work environment considered. An implicit objective of the study is to have a single characterization by sound work environment (sector or industry), so as not to carry out too many characterization phases. Other input SNRs for the characterization phase have been tested beforehand and the one equal to 0 dB has been retained because it has made it possible to achieve a characterization of the hearing aid allowing to properly simulate its behavior on the dynamics of −6 dB(A) to + 6 dB(A) in the sound environments studied. Regarding the simulation phase, 12 sound scenarios were created to simulate the acoustic conditions of an office and 12 others to recreate work conditions in an industrial environment with SNRs ranging from −6 dB(A) to +6 dB(A) by changing the noise level from 51 to 63 dB(A) in 3 dB(A) steps for the service sector and from 63 to 75 dB(A) in 3 dB(A) steps for the industrial sector. To create these auditory scenes, four speech signals (two male voices and two female voices) and three masking signals were selected to represent the service sector environment (speech noise and office noises) and three other masking signals (noisy machines) to represent the industrial environment. The Combescure sentences in French spoken by a man and a woman were used as vocal material as were two literary interviews broadcast by a French public radio station and downloaded from its website. The noise signals were built with the Audacity software using sample taken from an online sound library (https://lasonotheque.org/).
Hearing aid characterization results
Firstly, for the two types of sound environment (service sector and industrial sector), the nonlinear behavior of the hearing aids was studied for the two programs implemented, P1, P3, and the linear behavior for the P2 program. As presented in paragraph 2.1, the nonlinear frequency behavior of a system can be studied by analyzing Hn filters using the SSS method.
The study of all the Hn filters (not shown here) revealed that the hearing aids all had similar frequency responses when using the same program. As an example, the frequency responses Hn (for n=1, 2, 3) resulting from each characterization of hearing HA-A are shown in [Figure 5].
|Figure 5 Frequency responses of Hn filters of hearing aid (HA-A) for characterizations: A) P1 at 57 (service sector) dB(A), B) P2 at 57 dB(A), C) P3 at 57 dB(A), D) P1 at 69 dB(A) (industrial sector), E) P2 at 69 dB(A), and F) P3 at 69 dB(A)|
Click here to view
[Figure 5]A shows that the amplification of the hearing aids is between 50 Hz and 8 kHz, with a maximum amplification of 45 dB around a frequency of 4 kHz. The study of the higher order Hn filters (n≥2) resulting from the hearing aid’s characterization for a service sector environment using the P1 program showed that the behavior of the hearing aid can be considered as linear. Indeed, the Hn (n≥2) filters had a frequency response with much lower amplitude than the linear filter of order 1 (H1): about 40 dB lower over the entire spectrum. The analysis of the Hn filters resulting from the characterization for an industrial work environment [Figure 5]D also showed that the behavior of the hearing aid using program P1 was almost linear.
[Figure 5]B and [Figure 5]E] illustrate the characterization of the hearing aid when noise reduction is active (P2 program) for the service and the industrial sectors, respectively. These two figures show the transfer function of the noise which describes the processing that the hearing aid performs on the noise signal, and the filter of order 1 (H1) for the characterization of useful signals. Indeed, as indicated previously, the hearing aid’s behavior is considered linear in the characterization phase. As with the P1 program, there is a linear amplification of the H1 filter of the hearing aids of 40 dB around 4 kHz. The noise is amplified by the hearing aid, but less than the speech signal between 1 kHz and 2.5 kHz. Outside this frequency range, the noise transfer function and the H1 filter have almost the same response. This frequency band is crucial for understanding speech, and the hearing aid provides an improved SNR between 1 kHz and 2.5 kHz.
[Figure 5]C and [Figure 5]F present the Hn filters related to the characterizations performed for the service and industrial sector sound environments, respectively with the P3 program. It can be seen once again that the frequency behavior of the hearing aid can be considered as linear. The linear amplification (H1 filter) reached values up to 30 dB, contrary to the previous cases where the amplification reached 40 dB.
In order to check the hypothesis of linearity for the P2 program, the filter H1 is compared with the linear transfer function of the hearing aid [Figure 6]. This linear transfer function was obtained from the ratio between the SSS′ and SSS spectra.
|Figure 6 Comparison between the H1 filter obtained with the SSS method and the transfer function obtained with a linear method at 57 dB(A) (left) and at 69 dB(A) (right)|
Click here to view
There is almost no difference between the two curves in both cases, suggesting that the hypothesis of linearity for the P2 program is plausible.
| Simulation stage|| |
Once the hearing aids are characterized, it is possible to simulate speech and noise signals at their outputs. The way how these realistic signals are obtained will be described later in the paper. It is also possible to use the input speech and noise signals to estimate the STI (and the intelligibility scores) and compare them with the STI and intelligibility scores measured at the output of the hearing aid. The STI and the intelligibility scores are defined in the IEC 60268-16 standard. The STI represents the index of speech transmission via a physical system (e.g., a room) and/or an electroacoustic system (such as a hearing aid). It is based on the idea that the modulation of speech is deteriorated through such a system. This deterioration is estimated using the modulation transfer function, which depends on the SNR between the useful and the nonuseful signal (noise), the frequency masking and the hearing threshold. In order to consider hearing losses due to deafness in the calculation of the index, the thresholds of hearing perception defined in the standard were increased according to the hearing losses. The hearing losses in the octave bands ranging from 125 Hz and 8 kHz are 10, 25, 30, 40, 50, 55, and 65 dB HL, respectively.
The experimental protocol described in this section is applied to three selected brands of hearing aids and for the three signal processing programs implemented in each of them.
Different sound environments were created. These environments and the signals used (speech and noise signals) are described in paragraph 2.3.3.
The simulation stage is presented in [Figure 7]. For each sound environment, the following signals are emitted successively: the sum of the speech and noise signals and the sum of the speech and noise signals in phase inversion. The signals are recorded at the input of the hearing aid via a microphone located very close to the microphone system installed in the hearing aids. The Hagerman and Olofsson method is then used to separate noise from speech signals at the hearing aid input (Noise IN and Speech IN). When the noise reduction feature is not active, the Noise IN and the Speech IN signals are used to solve the Hammerstein model (Noise SIM and Speech SIM). In the case where the noise reduction feature is active, the noise transfer function and the H1 filter are used to calculate the Noise SIM and Speech SIM signals. The simulated speech and noise signals allow calculating the SNR and thus estimating the value of the STI at the output of the hearing aid called STI SIM.
|Figure 7 Block diagram of the simulation protocol (NR = Noise reduction)|
Click here to view
In order to check the validity of the simulation, the phase-inversion method is also applied to the signals measured at the output of the hearing aid in order to isolate the speech (Speech OUT) from the noise (Noise OUT) and to estimate the STI at the output of the hearing aid (STI OUT).
In order to assess the variations of intelligibility due to the hearing aid and the relevance of the simulation, the following statistical analyses were performed consecutively:
- One-way repeated measures ANOVA with a significance threshold α of 0.05 were performed with the JASP software, version 0.16.3.0, to compare the STI OUT values and the STI SIM values as a function of the SNR at the input of the hearing aid.
- One-way repeated measures ANOVA were also performed between the STI OUT and STI IN values to conclude on the benefit provided by the hearing aid.
If the p value is lower than the significance threshold α=0.05, the difference between the series is considered as significant. In [Figure 8],[Figure 9],[Figure 10],[Figure 11],[Figure 12],[Figure 13], one or more stars are displayed to indicate the significance of the comparison: no star if the difference is not significant, one if 0.05>p>0.001, two if 0.01>p>0.001, and three stars when 0.001>p.
|Figure 8 Comparison of STI values (IN − OUT − SIM) as a function of the input SNR all hearing aids confounded, using program P1 in a service sector work environment|
Click here to view
|Figure 9 Comparison of STI values (IN − OUT − SIM) as a function of the input SNR, all hearing aids confounded, using program P1 in an industrial sector work environment|
Click here to view
|Figure 10 Comparison of STI values (IN − OUT − SIM) as a function of the input SNR, all hearing aids confounded, using program P2 in a service sector work environment|
Click here to view
|Figure 11 Comparison of STI values (IN − OUT − SIM) as a function of the input SNR, all hearing aids confounded, using program P2 in an industrial sector work environment|
Click here to view
|Figure 12 Comparison of STI values (IN − OUT − SIM) as a function of the input SNR, all hearing aids confounded, using program P3 in a service sector work environment|
Click here to view
|Figure 13 Comparison of STI values (IN − OUT − SIM) as a function of the input SNR, all hearing aids confounded, using program P3 in an industrial work environment|
Click here to view
In order to have a sufficient number of values to perform these statistical tests, the results obtained for each brand of hearing aid were grouped. The STI values obtained for each brand of hearing aid were gathered by SNR for each combination of speech and noise signal. There are three brands of hearing aid, four speech signals, and three noise signals (i.e., 3 × 3 × 4 = 36) by environment considered.
Assessment of intelligibility
Following the methodology described in section 2.2, the STI values calculated at the output of the hearing aid (STI OUT) were compared to both the STI values calculated at the input of the hearing aids (STI IN) and to the STI values simulated at the output of the hearing aids (STI SIM). In each of [Figure 8],[Figure 9],[Figure 10],[Figure 11],[Figure 12],[Figure 13], the average values of the series of STI by SNR at the input of all the hearing aids are shown. The stars represented between the STI IN and the STI OUT values show the significance between the two series. In the same way, the stars represented between the STI OUT and the STI SIM values show the significance between these two series.
[Figure 9] present the STI results for program P1 in service and industrial work environments, respectively. The STI SIM/OUT comparisons of program P1 for the two work environments considered show that there is no significant difference between the measures and the simulations, thus illustrating that the characterizations led to a good estimation of STI at the output of the hearing aid.
In a service sector work environment, hearing aid program P1 provided a statistically significant (p≤0.001) increase in STI values for all the input SNRs. The STI IN values in the service setting were low due to taking hearing losses into account in the STI calculation. Since the acoustic level of the different sound scenarios was around 57 dB(A) in a context where the masking noise covered the frequency range of speech, the speech modulation transfer functions were mostly equal to 0. When the hearing aids are used, the amplification provided by the devices [Figure 5]A allowed the speech modulation transfer function to exceed 0, thus increasing the STI value at the output of the hearing aid.
In an industrial work setting, hearing aids provide only a statistically significant increase in STI values for positive or zero input SNRs [cf. Figure 9]. The STI IN values were less clustered than those seen in a service setting, since the hearing aid input levels were higher.
In both work environments, for all the input SNRs studied, there was a statistically significant increase in intelligibility scores which were also simulated accurately.
[Figure 10] and [Figure 11] present the STI results when program P2 was activated in the service sector and industrial sector work environments, respectively. In both cases, the characterizations of the hearing aids made it possible to obtain a good estimation of the STI SIM at the output of the hearing aid. In addition, all the p values comparing the IN/OUT STI were lower than 0.001 for all the environments considered (except for an input SNR of 6 dB(A) in an industrial environment where 0.01>p>0.001).
[Figure 12] and [Figure 13] present the STI results when the P3 program is considered for the service and industrial work environments, respectively. The comparisons of STI SIM/OUT showed that the characterizations of hearing aids did not permit a good estimation of intelligibility at the output of the hearing aid in a service sector work environment. Indeed, for this case, the differences between the STI SIM/OUT values were all very significant (p≤0.001. In an industrial work environment, the STI values were estimated better. The comparisons of the STI IN/OUT values show that there was a significant hearing aid effect on speech intelligibility in both work environments for all the input SNRs.
Lastly, [Figure 14] and [Figure 15] show the average differences between IN/OUT intelligibility scores as a function of program for each input SNR considered in order to directly compare the efficiency of the hearing aids to improve speech intelligibility in noise. The intelligibility scores were estimated on the basis of the set of curves given in annex. E of standard IEC 60268-16. In this annexe., the curve corresponding to the intelligibility score obtained for sentences as vocal material is used since sentences are used as speech signal.
|Figure 14 Comparison of efficiency between the three programs, P1, P2, and P3 in terms of improving the speech intelligibility score in a service work environment as a function of the input SNR|
Click here to view
|Figure 15 Comparison of efficiency between the three programs, P1, P2, and P3 in terms of improving the intelligibility score in an industrial work environment as a function of the input SNR|
Click here to view
| Discussion & Conclusion|| |
In this study, an original protocol was implemented to characterize a hearing aid by taking into account possible frequency nonlinearities in order to estimate intelligibility at the output of the hearing aid in different noisy work environments. This protocol was based on the use of the experimental SSS methods for the two programs where noise reduction was not used, P1 and P3.
For the program where noise reduction was active, a linear approach was proposed for the characterization, with the joint use of the SSS method and the phase-inversion method. The frequency linearity hypothesis was checked.
The characterizations of hearing aids using programs P1 and P3 (with compression) did not highlight the presence of frequency nonlinearities since all the Hn filters of an order higher than 1 were of negligible amplitude compared with filter H1. The hearing aid characterizations enabled estimating the STI values and the intelligibility scores at the output of the hearing aids for the two types of sound environments studied and for each input SNR when they used programs P1 and P2. However, the experimental protocol was not adapted to estimate speech intelligibility when the hearing aids used program P3 containing the compression. Indeed, the characterization stage did not permit taking into account the effect of this functionality since all the Hn filters are linear and the characterization was performed at a fixed level. The compression implemented in program P3 performs an amplification in the different frequency channels as a function of the acoustic level perceived in these channels at the input of the hearing aid. The speech and noise signal levels varied with time and were not necessarily amplified with the same gain as the signals used during the characterization, which perhaps explains the poor estimation of the SNR values and thus the STI values. Noise in an industrial environment is more stationary than in an office environment and is therefore much less sensitive to compression, so simulating hearing aids may make it possible to estimate intelligibility in the industrial sector accurately.
It is therefore possible to use the experimental protocols to estimate speech intelligibility in noise for a hearing-impaired employee in their workplace for classical amplification programs (P1) and for those comprising noise reduction options (P2). It will therefore be possible to better understand their situation and propose, if necessary, treating work premises or adapting workstations for these two processing options. Although the characterizations did not highlight any frequency nonlinearities in the P1 and P3 programs studied here, other programs or other presetting methods (such as those recommended by the manufacturers) could present nonlinearities which would be taken into account with this method.
In the two work environments studied, the noise reduction functionality in program P2 led to improvements of speech intelligibility in noise. Programs P1 and P3 also provide a significant improvement in noise intelligibility in both environments, but program P2 is much more effective in an industrial work environment. In all the work sound environments studied in which hearing-impaired employees wear hearing aids, all the programs considerably improved speech intelligibility in noise with an improvement of at least 15% of the intelligibility score in the worst case.
Globally, all the hearing aid programs provided a very significant improvement in intelligibility in a service work context due to the amplification of acoustic signals which offsets hearing loss when the overall input level is around 57 dB(A). There was a better improvement in intelligibility with program P2 due to the noise reduction feature in the service environment and in the industrial one than when using programs P1 and P3.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Hétu R. La surdité professionnelle: un handicap sensoriel trop discret [Professional deafness: a discreet sensory handicap.]. Sante Ment Que 1981;6:155–63.
Deshaies P, Martin R, Belzile D et al.
Noise as an explanatory factor in work-related fatality reports. Noise Health 2015;17:294–9.
] [Full text]
Hétu R, Getty L, Waridel S. Attitudes towards co-workers affected by occupational hearing loss. II: Focus groups interviews. Br J Audiol 1994;28:313–25.
Laplante-Lévesque A, Hickson L, Worrall L. Factors influencing rehabilitation decisions of adults with acquired hearing impairment. Int J Audiol 2010;49:497–507.
Renard C, Ménard M. Technical developments in hearing aids (in French). Acoust Techniq 2009; 56:24–35.
Brons I, Houben R, Dreschler WA. Effects of noise reduction on speech intelligibility, perceived listening effort, and personal preference in hearing-impaired listeners. Trends Hear 2014;18:1–10.
Leroux N, Laroche C, Giguère C, Voix J. Hearing aid use in noisy workplace, Document published by the Robert-Sauvé Research Institute in Occupational Health and Safety (IRSST), Studies and research projects. 2018; R-1015.
Houtgast T, Steeneken H. Past, Present and Future of the Speech Transmission Index, The Netherlands: TNO Human Factors, 2002. p. 18–25.
Tan AH, Godfrey KR. The generation of binary and near-binary pseudo-random signals. Proceedings of the 18th IEEE Instrumentation and Measurement Technology Conference 2001;2:583–8.
Blümich B. White noise nonlinear system analysis in nuclear magnetic resonance spectroscopy. Progress in Nuclear Magnetic Resonance Spectroscopy 1987;19:331–17.
Solomou M, Rees D, Chiras N. Frequency domain analysis of nonlinear systems driven by multiharmonic signals. IEEE Transactions on Instrumentation and Measurement 2004;53:243–0.
Schoukens J, Pintelon R, Rolain Y, Dobrowiecki T. Frequency response function measurements in the presence of nonlinear distortions. Automatica 2001;37:939–6.
Crama P, Schoukens J. Initial estimates of Wiener and Hammerstein systems using multisine excitation. IEEE transactions on Instrumentation and Measurement 2001;50:1791–5.
Farina A. Simultaneous measurement of impulse response and distortion with a swept-sine technique. AES 108th convention; 2000.
Novàk A, Simon L. Nonlinear System Identification Using Exponential Swept-Sine Signal. IEEE Transactions on Instrumentation and Measurement 2009;59:1–9.
Novàk A, Lotton P, Simon L. Synchronized Swept Sine: theory, application and implementation. J AES 2015;63:786–8.
Hagerman B, Olofsson A. A method to measure the effect of noise reduction algorithms using simultaneous speech and noise. Acta Acustica 2004;90:356–1.
Schädler M, Warzybok A, Kollmeier B. objective prediction of hearing aid benefit across listener groups using machine learning: speech recognition performance with binaural noise-reduction algorithms. Trend Hear 2018;22:1–21.
Ephraïm Y, Malah D. Speech enhancement using a minimum meansquare error log-spectral amplitude estimator. IEEE Trans. on Acoustics, Speech and Signal Processing 1985;33:443–5.
Miller CW, Bentler RA, Wu YH, Lewis J, Tremblay K. Output signal-to-noise ratio and speech perception in noise: effects of algorithm. Int J Audiol 2017;56:568–79.
Polette O. Comparative study of the two pre-selection formulas NAL-NL2 and DSL version 5 on different types of hearing loss (in French). 2012.
Holube I, Fredelake S, Vlaming M, Kollmeier B. Development and analysis of an international speech test signal (ISTS). Int J Audiol 2010;49:891–03.
Combescure P, Le Guyader A, Gilloire A. Quality evaluation of 32 kbit/s coded speech by means of degradation category ratings 1982. IEEE International Conference on Acoustics, Speech, and Signal Processing; 7:988–1.
International Electrotechnical Commission. Sound system equipment: Objective rating of speech intelligibility by speech transmission index (IEC 60268-16:2020), 2020.
Institut National de Recherche et Sécurité, 1 Rue du Morvan CS 60027, 54519 Vandœuvre-lès-Nancy;
Source of Support: None, Conflict of Interest: None
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5], [Figure 6], [Figure 7], [Figure 8], [Figure 9], [Figure 10], [Figure 11], [Figure 12], [Figure 13], [Figure 14], [Figure 15]