Method and system for controlling noise

Abstract

A method for controlling noise for a human subject, comprising: the human subject receiving a sound signal through a headphone, the sound signal comprising a noise signal; generating a cancellation signal based on a hearing target curve related to sound amplitude and frequency; and applying the cancellation signal to the sound signal such that the noise signal is attenuated.

Description

Technical Field

[0001] The present disclosure generally relates to noise control related to sound signals received in the human ear. Background Art

[0002] Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by impairments in social interaction and communication, as well as repetitive, restricted, and stereotyped patterns of behavior. It is accompanied by various sensory characteristics, such as hyper- or hypo-responsiveness to sensory input, etc. When autistic individuals who are sensitive to sound perceive unpleasant auditory stimuli, it not only causes strong reactions, but also leads to reduced participation in important life activities and avoidance of specific environments and interactions. Existing environmental noise control methods include controlling the sound propagation path, such as by installing sound-absorbing panels, and by erecting barriers to demarcate quiet areas. Existing systems and methods also include earmuffs and portable noise-canceling headphones for directly controlling noise at the receiver. They provide a barrier between the ear and the external environment.

[0003] The purpose of the present disclosure is to overcome or significantly improve one or more disadvantages of the prior art, or at least provide a useful alternative. Summary of the Invention

[0004] In one aspect of the present disclosure, a method for controlling noise for a human subject is provided. The method includes: a human subject receiving a sound signal through a headset, the sound signal including a noise signal; generating a cancellation signal based on a hearing target curve related to sound amplitude and frequency; and applying the cancellation signal to the sound signal such that the noise signal is attenuated.

[0005] Additionally or alternatively, the hearing target curve includes the relationship between noise attenuation in decibels (dB) and sound frequency. The sound frequency can be in the range of 250 Hz to 8000 Hz.

[0006] Additionally or alternatively, the method further includes: determining a sound intensity hearing level corresponding to the neutral response of the human subject based on an auditory perception curve; and calculating the noise attenuation based on the difference between the sound intensity hearing level and the noise level.

[0007] Additionally or alternatively, the method further includes determining the neutral response by performing power function curve fitting where y i is the i-th average perception level, x i is the i-th intensity level, and a, b, c are coefficients.

[0008] Additionally or alternatively, the method further includes performing a clustering algorithm on multiple human subjects. The clustering algorithm can include an agglomerative hierarchical algorithm.

[0009] Additionally or alternatively, the method includes performing electroencephalogram (EEG) tests on multiple human subjects to obtain the neural responses of the multiple human subjects in response to sound stimuli.

[0010] Additionally or alternatively, the method further includes: recording the data measured by the EEG tests; rereferencing the data to obtain rereferenced data; filtering the rereferenced data to obtain filtered data; and determining a search window to identify a first peak P1, a second peak P2, and a trough N1.

[0011] Additionally or alternatively, the method includes performing baseline correction on the filtered data. The method may further include averaging the filtered data within the search window.

[0012] Additionally or alternatively, the method further includes: configuring a sound signal such that the sound signal propagates along a main path and evolves into a residual signal; and performing superposition of the residual signal and a cancellation signal to generate an error signal.

[0013] Additionally or alternatively, the method further includes: measuring the error signal; calculating the difference between the sound signal and the error signal; and comparing the difference and the cancellation signal. If the difference and the cancellation signal are greater than a threshold, the method adjusts at least one parameter of the adaptive filter.

[0014] In another aspect of the present disclosure, a system for controlling noise for a human subject is provided. The system includes a headset configured to receive a sound signal and a computer device. The computer device is configured to generate a cancellation signal based on a hearing target curve related to sound amplitude and frequency and apply the cancellation signal to the sound signal such that the noise signal is attenuated.

[0015] Additionally or alternatively, the system further includes: a main path configured to transmit the sound signal such that the sound signal evolves into a residual signal; and a secondary path configured to transmit the cancellation signal; performing a superposition operation on the residual signal and the cancellation signal at the intersection of the main path and the secondary path to generate an error signal. An adaptive filter may be provided on the secondary path.

[0016] Additionally or alternatively, the headset includes a reference microphone configured to measure the sound signal; and an error microphone configured to measure the error signal.

[0017] Additionally or alternatively, the computer device calculates the difference between the sound signal and the error signal and compares the difference with the cancellation signal. When the difference and the cancellation signal are greater than a threshold, the computer device may further adjust at least one parameter of the adaptive filter.

[0018] Other example embodiments are discussed herein. Description of the Drawings

[0019] Embodiments of the present disclosure will now be described with reference to the accompanying drawings by way of example only, in which:

[0020] Figure 1A An example of a tonal sound with a duration of 1 second (s) and a start / end ramp of 20 milliseconds (ms) is shown, according to certain embodiments of the present disclosure.

[0021] Figure 1B Calibration settings of a sound stimulation presentation system are shown, according to certain embodiments of the present disclosure.

[0022] Figure 2 A Likert five-point scale and its corresponding emojis used in an auditory perception test are shown, according to certain embodiments of the present disclosure.

[0023] Figure 3A In accordance with certain embodiments of the present invention, participants in an electroencephalogram test (EEG) wear an EEG cap and record raw EEG signals while listening to sound stimuli through headphones.

[0024] Figure 3B Shows Figure 3A EEG electrode positions in an EEG test.

[0025] Figure 3C Shows Figure 3A The experimental setup of an EEG test.

[0026] Figure 3D Shows how recorded EEG data is processed, according to certain embodiments of the present disclosure.

[0027] Figure 4 Average scores of auditory perception grouped by different sound intensity hearing levels are shown: (a) 30 dB HL; (b) 40 dB HL; (c) 50 dB HL; (d) 60 dB HL; (e) 70 dB HL; (f) 78 dB HL.

[0028] Figure 5 A comparison of (a) peak amplitude and (b) peak latency of the N1 component at channel T8 in a typical development (TD) group and an ASD group is shown.

[0029] Figure 6 A block diagram of the K-means clustering algorithm is shown, according to certain embodiments of the present disclosure.

[0030] Figure 7 Results of clustering validation indices using three clustering methods with K ranging from 3 to 6 are shown: (a) Silhouette index; (b) Calinski-Harabasz index; (c) Davies-Bouldin index.

[0031] Figure 8 Shows the average scores of auditory perception for five clusters of the K - means algorithm at the following different sound intensity hearing levels: (a) 30 dB HL; (b) 40 dB HL; (c) 50 dB HL; (d) 60 dB HL; (e) 70 dB HL; (f) 78 dB HL.

[0032] Figure 9A Shows the auditory perception curves, along with the auditory perception responses at the following different frequencies: (i) 250 Hz; (ii) 500 Hz; (iii) 1 kHz; (iv) 2 kHz; (v) 4 kHz; (vi) 8 kHz.

[0033] Figure 9B Shows the Figure 9A result target curve of noise cancellation obtained from

[0034] Figure 10A Shows a block diagram of a proposed active noise cancellation (ANC) system according to certain embodiments of the present disclosure.

[0035] Figure 10B Shows from another perspective the Figure 10A system of

[0036] Figure 11 Shows the comparison of auditory perception responses with and without ANC at the following different frequencies: (a) 250 Hz; (b) 500 Hz; (c) 1 kHz; (d) 2 kHz; (e) 4 kHz; (f) 8 kHz. Detailed Description

[0037] The present disclosure will now be described with reference to the following examples, which should be considered illustrative and non - limiting in all respects. In the figures, corresponding features in the same embodiments or features common to different embodiments are given the same or similar reference numerals.

[0038] Throughout the specification and claims, the words "comprising", "including", etc. shall be construed in an inclusive sense, rather than an exclusive or exhaustive sense; that is, in the sense of "including, but not limited to".

[0039] Furthermore, as used herein and unless otherwise specified, the use of ordinal adjectives "first", "second", etc. to describe a common object only indicates different instances of the similar objects being referred to, and is not intended to mean that the objects described must be in a given order, whether in terms of time, space, ranking, or any other way.

[0040] Example embodiments relate to methods and systems for controlling noise, taking into account the individual auditory perception of a human subject. The human subject described herein can be a person with autism or hearing problems, or a normal person who wishes to avoid accidental or unwanted noise. Although, in the following, embodiments of the present disclosure may be described with reference to a person with autism, it will be recognized that the present disclosure is applicable to persons without autism or other physical or mental problems.

[0041] Many existing systems or methods are defective in one or more aspects. For example, some systems or methods use fixed noise suppression facilities at one location, which is not suitable for people who move around in their daily lives and engage in activities at different locations. The prior art does not consider the effects of the physical characteristics of sound (including frequency and sound intensity level). Many existing systems perform poorly in the low-frequency region. Some techniques are based only on the suppression of sound pressure level and use the same type of noise cancellation function for all human subjects without considering their individual auditory perception.

[0042] Example embodiments solve one or more of these problems associated with the prior art and provide a technical solution with a novel design.

[0043] According to one or more embodiments, in order to design a suitable noise control function in a headset to meet the needs of people with autism and different auditory perceptions, a series of auditory perception and electroencephalogram (EEG) tests were conducted, in which autistic participants with auditory over-responsiveness listened to sounds of different frequencies and amplitudes so that their subjective auditory responses could be analyzed. A suitable noise attenuation target curve or hearing target curve was determined based on the auditory perception curve, which was constructed using a power function fit as a function of the average auditory perception level and the noise level. Subsequently, a hybrid active noise cancellation (ANC) system based on auditory perception was developed and validated. The results showed that most autistic participants rated the frequencies of 250 Hz and 8 kHz as the most unpleasant. For example, the K-means algorithm was used to divide the participants into five clusters. It was found that each cluster had its own unique auditory perception response. Finally, when the participants used this headset or headphones with auditory perception characteristics suitable for different autistic participant clusters, an improvement in the auditory perception response was observed. Although the participants described herein are children or adolescents, it will be understood that they are merely examples of human subjects and are for illustrative purposes only. In some embodiments, the participants can be people of other age groups.

[0044] According to one or more embodiments, subjective auditory perception and EEG tests of different sound stimuli are conducted on participants with autism and normal development, so that the auditory perception responses and characteristics of these participants to sound stimuli of various frequencies and amplitudes can be understood and quantified. There is a correlation between the auditory perception level and the amplitude of the slow-wave cortical auditory evoked potential. Generally, autistic participants in all clusters feel unpleasant, especially at 250 Hz and 8 kHz, although the obtained perception levels vary according to the noise level. Different clusters have their own frequency and sound intensity hearing level response characteristics to sound stimuli. This indicates that the need for noise control at specific frequencies causes annoyance to different participants to varying degrees. An active noise control system in a headset or headphones with an auditory perception response function is developed to alleviate the adverse auditory behaviors of autistic participants, and its performance and improvement are verified through experiments and investigations.

[0045] In order to design a suitable noise control function in headphones to cater to autistic participants with different auditory perceptions, one or more embodiments of the present disclosure investigate the auditory responses of autistic and auditory over-responsive participants in terms of amplitude and frequency, establish an evaluation method that can quantify the perception of sound by autistic participants, determine the relationship between the physical parameters of sound and the subjective auditory response, and develop a suitable human perception ANC method to effectively alleviate the adverse behaviors related to auditory over-responsiveness of autistic participants.

[0046] To illustrate the creative concept and show how the creative concept is implemented, experiments will be described and discussed below according to certain embodiments. In these experiments, children or adolescents with or without autism are participants. It will be understood that the experiments, including various numerical values, numbers, etc. used or selected, are for illustrative purposes only and should not be construed as limiting.

[0047] For example, in order to understand the auditory perception of autistic participants and the differences between participants with and without autism, the evaluation of their sound responses is conducted in two sessions. The first session focuses on the subjective evaluation experiment, which directly reflects the subjective auditory perception or response to different sound stimuli or excitations; the second session includes the physiological auditory response, which reflects the intermediate neural response to sound excitation and its corresponding emotions.

[0048] There were two groups of participants: typically developing participants (TD) and participants with autism (ASD), where participants in the TD group did not have autism. A total of 83 ASD participants (75 males and 8 females, with a mean age of 9 ± 1.7 years) and 50 TD participants (38 males and 19 females, with a mean age of 10 ± 1.4 years) were recruited through purposive sampling and snowball sampling. The recruited participants with autism were diagnosed with autism, autistic disorder, or Asperger's syndrome and completed the Hong Kong Chinese version of the Autism Spectrum Scale, aged 7 - 12 years and with primary school education. The participants were able to give oral responses using a Likert five - point scale. The normal hearing function of these participants was evaluated using a pure - tone audiometry hearing test. For the hearing test, all participants underwent two hearing threshold screenings at frequencies of 250 Hz, 500 Hz, 1000 Hz, 2000 Hz, 3000 Hz, 4000 Hz, and 8000 Hz, with three different sound intensity hearing levels (10, 15, and 20 dB HL). The participants were asked to indicate verbally or by gesture whether they could hear the sound emitted from the headphones. The average hearing level of the participants at all measured frequencies was higher than the acceptable level of 15 dB HL. In addition, these participants scored 85 or higher on the Test of Non - Verbal Intelligence - Fourth Edition (TONI - 4). To obtain their neural responses to sound or acoustic stimuli, the participants were confirmed to have no neurological diseases. The participants with autism also completed an auditory over - reactivity screening using the Chinese version of the Sensory Profile or Auditory Over - Reactivity, whereby a score of 30 or lower was defined as having auditory over - reactivity.

[0049] To obtain the auditory perception and auditory responses of the two groups of participants, the sound stimuli were focused on tone signals with different frequencies and amplitudes. The full set of sound stimuli included 36 channels, with six different frequencies (0.25 kHz, 0.5 kHz, 1 kHz, 2 kHz, 4 kHz, and 8 kHz) and six different sound intensity hearing levels (30, 40, 50, 60, 70, and 78 dB HL), where dBHL is the decibel of the hearing level commonly used in audiology, and 0 dB HL is the average hearing threshold of normal - hearing ordinary listeners, in units of dB sound pressure level. These six center octave frequencies covered almost the entire frequency range of environmental sounds in the community. Each tone sound with the corresponding amplitude was generated to last for 1 second, with a start / end ramp of 20 milliseconds, as Figure 1A shown. The full set of sound stimuli with the above - mentioned frequencies and amplitudes was repeated three times. The device used to present the sound stimuli was calibrated using a head and torso simulator to ensure accurate transmission of the sound stimuli. The calibration settings are as Figure 1B shown, which shows the head and torso simulator 102 and the signal conditioner 104.

[0050] The subjective auditory perception or response of the participants was evaluated in a soundproof chamber. During the experiment, sound stimuli were played using a computer connected to Bose QC35II headphones with an audio amplifier. The experimental control software E-Prime 2.0 was used to create a random sequence of sound stimuli for each participant. This enabled the researchers to record the participants' responses using a response recorder in the form of a response pad without them knowing the sequence of sound stimuli in advance. Additionally, the software allowed the researchers to insert stimulus intervals of different durations (time intervals of silence in this experiment) based on the participants' responses after each sound stimulus. The procedure for presenting each sound stimulus was as follows. Before presenting each sound stimulus, a black fixation cross appeared in the center of the screen to attract the participants' attention. After the sound was played, a Likert five-point scale along with corresponding emojis, such as Figure 2 as shown. The Likert scale was designed as a bipolar scale to capture both the pleasant and unpleasant feelings of the participants when listening to the sound stimuli. The participants were asked to verbally rate how much they liked or disliked the sound. A rating of +2 indicated "liked very much" and was accompanied by an emoji of a big laugh; +1 indicated "liked" and was accompanied by a smiling face emoji; 0 indicated "neutral" and was accompanied by a neutral emoji; -1 indicated "disliked" and was accompanied by a sad emoji; -2 indicated "disliked very much" and was accompanied by a very sad emoji on the monitor. If the participants did not hear the sound, the researchers repeated the sound. The stimulus intervals varied between 2 and 10 seconds, depending on the participants' responses, to avoid habituation effects. For trials with a rating of -1 (disliked) or -2 (disliked very much), the intervals varied between 8 and 10 seconds. For trials with a rating of 0 (neutral), the interval was 5 seconds. For trials with a rating of +1 (liked) or +2 (liked very much), the intervals varied between 2 and 4 seconds. The entire set of sound stimuli consisted of 36 sound stimuli, which were randomly repeated three times to check reproducibility. The total duration of the entire experiment was approximately 30 minutes (min).

[0051] Reference Figure 3A 、 3B 、3C and 3D, to examine the reliability and consistency of the participants' sound perception responses to the sound stimuli, the participants' neural responses to the sound stimuli were measured by electroencephalogram (EEG) tests. The experimental setup was as Figure 3CAs shown. The sound generator 310 (e.g., a computer with sound generation software) generates a sound or sound signal, which is input into the headphones 320 that can be worn on the head of the participant. The sound signal stimulates the participant's brain, and the electrical activity in the brain is measured by the EEG 330 through small metal electrodes attached to the participant's scalp. Then, the measured data is input and processed by the computer device 340. The computer device 340 includes a processor 342, a memory 344, and a software application 346. When executed by the processor 342, the software application 346 (e.g., a MATLAB program) causes the processor 342 to execute one or more steps to process the measured data, as shown below.

[0052] In this embodiment, the sound stimuli in this embodiment are concentrated on 18 channels including the same frequencies (250 Hz, 500 Hz, 1000 Hz, 2000 Hz, 4000 Hz, and 8000 Hz) as in the above-mentioned auditory perception test and three different sound intensity hearing levels of 40, 60, and 78 dB HL. The number of sound stimuli in the EEG test is less than that in the sound perception response session because the required time cannot be tolerated by autistic participants for too long. All sound stimuli are generated by Panasonic RP-HD5 headphones controlled by E-Prime 2.0 software, with a duration of 200 ms and a start / end ramp of 20 ms. In this experiment, the participant sits on a comfortable chair in a soundproof, electrically shielded, and dimly lit room, as Figure 3A shown. During the entire experiment, the participant is instructed not to focus on the sound but to watch a silent movie of their choice. They are required to stay still and try to reduce the blink frequency. As Figure 3B shown, three EEG electrodes (denoted as Fz, Cz, and Pz) are placed at the frontal, central, and parietal positions along the midline sagittal plane of the head respectively. The other two electrodes are located on the left and right temples of the head, denoted as T7 and T8 respectively. The brain wave signals captured by the electrodes placed on the waveguard TM EEG cap are recorded by the ANT-Neuro eego TM mylab amplifier. The electrode located at the left mastoid (M1) is regarded as the reference, and the frontopolar midline electrode is regarded as the ground. The electrode located at the right mastoid (M2) is recorded for rereferencing during the offline processing stage. Another four Ag / AgCl cup electrodes are placed near the eyes to monitor eye movements. The electrodes are made of sintered Ag / AgCl. The electrode impedance is kept below 5 kΩ. The EEG is sampled at a rate of 1 kHz during the entire session. Each sound stimulus is generated 40 times for each participant in a random pattern. The stimulus interval jitters between 2 seconds and 3 seconds. The entire test includes 720 trials and lasts for about 55 minutes.

[0053] Figure 3D An example is shown of how the computer device 340 processes the recorded EEG data (i.e., the measurement data of EEG 330) input into MATLAB for offline processing and analysis to study the relationship between electrophysiology and the response of auditory perception to auditory stimuli. The EEG data is rereferenced to the average of the left and right mastoids (box 301). This provides a symmetric reference that does not bias either hemisphere. The equation for the reference process is shown below, taking the channel Fz as an example. This reference process is repeated for each recorded EEG channel. Let V Fz V M1 and V M2 be the absolute voltages at positions Fz, M1, and M2, respectively. Let be the voltage at the Fz point after rereferencing.

[0054]

[0055] Box 302 represents data filtering. In this embodiment, a window sinc filter is used as a notch filter at 50 Hz to filter the rereferenced data, with a filter kernel length of 1650 points to eliminate line noise. The filter kernel is given by h[i]=sin(2πf ci) / iπ. Another windowed sinc filter with a cut-off frequency of 40 Hz and a filter kernel length of 3300 points is used as a low-pass filter to mitigate high-frequency noise, such as muscle artifacts. Then, a windowed sinc filter with a cut-off frequency of 1 Hz and a filter kernel length of 3300 points is used as a high-pass filter to minimize low-frequency noise that may be caused by body movement, improper skin electrode contact, and breathing. The pseudo-shadow subspace reconstruction method is used to remove and correct bad channels and noise segments. To study the electrophysiological responses in response to sound stimuli, specific time windows near the start of each sound stimulus were extracted from the continuous EEG recordings. These time windows are time-locked to the sound stimulus time and are called epochs. In this study, the continuous EEG data was divided into 600-ms epochs, 100 ms before the start of each stimulus and 500 ms after the start of each stimulus. The 100-ms time interval before the start of each stimulus (baseline period) was used to perform baseline correction (block 303), where the mean of the EEG data in these 100-ms pre-stimulus time intervals was calculated and then subtracted from each time point in the baseline period and the post-stimulus intervals of each epoch. Baseline correction was performed to reduce the effect of baseline differences between epochs, which are meaningless for interpretation and may bias the results of data analysis. Epochs with signal amplitudes exceeding ±90 μV in any channel were excluded. The epochs corresponding to the same sound stimulus were averaged (block 304) to average out the spontaneous background EEG activity (such as noise), making the time-locked EEG responses caused by the sound stimulus different from the background. This averaging process was repeated for each sound stimulus, and the resulting time-locked EEG responses were derived. To ensure the reliability of the results, only data with components within the period described by the conventional slow-wave cortical auditory evoked potential were included in the analysis. These components are characteristic deflections that occur around specific peak latencies, where the peak latency is measured using the start of the stimulus as a reference point (i.e., starting at 0 ms at the start of the stimulus). In this study, the first and second positive peaks are denoted as P1 and P2, with peak latencies of approximately 50 ms and in the range of 175 - 200 ms, respectively. The first trough, which is denoted as N1, is a prominent negative wave with a peak occurring at approximately 100 ms. The peak amplitudes and latencies of the P1, N1, and P2 components in the time signal of the event-related potential were used to quantify the participants' neural responses to the auditory stimuli. These response characteristics are mainly affected by the physical properties of the stimulus, such as the duration of the sound stimulus, the rise time (the time it takes for the sound signal to go from silence to peak amplitude), the sound intensity level, the inter-stimulus interval, and the stimulus characteristics. In block 305, to identify these three components, the peaks of P1 and P2 were searched for in the time periods of 20 - 120 ms and 150 - 250 ms, respectively. For the first trough, the N1 component, the search window was centered on 70 ms to 150 ms.In this analysis, peaks were defined as the data points with the maximum positive amplitude of P1 and P2 and the maximum negative amplitude of N1 within the search window. The peak amplitude value was measured as the average amplitude of the data within ±1 millisecond around the peak, which is the average of the peak and the data values 1 millisecond before and after the peak. The peak amplitudes and latencies of the P1, N1, and P2 components were studied.

[0056] To improve data quality, participants were included in the analysis only if they showed consistent and reliable responses in both the auditory perception and EEG tests. In the auditory perception test, a one-way random effects, absolute agreement, multi-measure intraclass correlation coefficient (ICC) was used to evaluate the consistency of participants in three repeated auditory perception test evaluations. Scores less than or equal to 0.38 were considered inconsistent and excluded from the data analysis. Additionally, participants with fewer than 540 epochs after data preprocessing and cleaning were also excluded. Thus, 33 ASD participants and 12 TD participants were excluded. A total of 50 ASD participants and 38 TD participants aged between 7 and 12 years were included in the analysis.

[0057] The scores of the auditory perception test responses were adjusted from (-2 to +2) to positive numbers from (1 to 5), where the adjusted scale rank 1 represents "very dislike" and rank 5 represents "very like". This scale adjustment was made to facilitate data analysis and interpretation of auditory perception and electrophysiological responses to sound stimuli. For each participant, the responses to each sound stimulus in the three repeated evaluations in the auditory perception test were averaged to obtain the average score. This yielded 36 average scores for each participant. These scores represent the individual variation patterns of each participant and will be adopted in the subsequent further analysis. Additionally, the average scores of all 36 sound stimuli were added to the total score of each participant. The total score measures the auditory perception of the participant, considering their responses to the 36 sound stimuli with a unified weight. A lower total score indicates a greater dislike of the sound stimuli, while a higher total score indicates a greater liking of the sound stimuli. The total scores of all TD participants were averaged to obtain the average total score and used as a cut-off point to divide the two groups of autistic participants. Participants with a total score higher than the TD cut-off point were classified into ASD group 1, while participants with a total score lower than the TD cut-off point were classified into ASD group 2.

[0058] Figure 4The average scores of two ASD groups and the TD group based on auditory perception experiments are given. As can be seen, except for the average score of ASD group 1 at 250 Hz, 78 dB HL (as shown by the dashed line), the average scores are higher than those of the TD group at all frequencies and sound intensity hearing levels. At lower sound intensity hearing levels, such as 30 and 40 dB HL, the average scores increase in the frequency range from 250 Hz to 1 kHz and then decrease towards the frequency of 8 kHz. This indicates that they dislike the frequencies of 250 Hz and 8 kHz more and generally prefer to listen to medium-frequency sounds of around 1 kHz. At 50 dB HL and 60 dB HL, the difference between the average scores at 250 Hz and 8 kHz and those of other frequencies becomes more obvious. At 70 dB HL, the curve of the average score changing with frequency is similar to that at 50 and 60 dB HL, and the average scores at almost all frequencies are lower than 3, indicating that they dislike these sounds. For most of the presented sound stimuli, the average score curve of ASD group 1 is significantly higher than that of the TD group participants. This result shows that ASD group 1 has a higher tolerance for high-intensity sounds. Compared with the TD group, ASD group 2 (as shown by the dashed line) has relatively lower average scores for all sound stimuli but shows a similar change pattern to the TD group. At 30 and 40 dB HL, the response is similar to the TD response, except that the average score at the frequency of 250 Hz is significantly lower. At 50 - 70 dB HL, this group of participants with autism shows an unpleasant response to all sound stimuli, and lower average scores are observed at all frequencies compared with the TD group.

[0059] The results from above showed that most autistic participants disliked very low and very high frequencies. To verify the subjective auditory perception, we investigated the relationship between this response and the EEG results. Spearman correlation analysis was used to analyze their correlation. The results of combined Group 1 and Group 2 of the ASD participants and the results of the TD participants are listed in Table 1 below. In the ASD group, there were significant correlations between auditory perception and the absolute N1 peak amplitude as well as the P1 and P2 peak latencies in specific EEG channels. The correlation coefficients for the N1 peak amplitude ranged from -0.118 to -0.149, p<0.01, and for the P1 and P2 peak latencies, the coefficients ranged from 0.122 to 0.194, p<0.01. In the TD group, there were significant correlations between auditory perception and the absolute N1 peak amplitude as well as the P1 and P2 peak latencies in multiple EEG channels, with the correlation coefficients ranging from -0.103 to -0.270, p<0.01 and from 0.109 to 0.170, p<0.01 respectively. Auditory perception is a subjective evaluation of the sounds provided by the participants, while the neural response shows an objective response to the sounds. The correlation between these two responses indicates that the average auditory perception score represents the subjective perception of the participants towards the presented sound stimuli. For both the ASD group and the TD group, the absolute N1 peak amplitude generally showed a better association with the auditory perception response. This suggests that the N1 peak amplitude might be a suitable candidate for quantifying the neural response of the participants to the sound stimuli. Generally, the higher the sound intensity hearing level, the higher the absolute peak amplitude of the N1 component and the lower the peak latency.

[0060] Table 1 Spearman correlations between event-related potential (ERP) component responses and ranks in the auditory perception test

[0061]

[0062]

[0063] *The correlation is significant at the 0.05 level (two-tailed).

[0064] **The correlation is significant at the 0.01 level (two-tailed).

[0065] Figure 5Shows the comparison of the peak amplitude and peak latency N1 component at channel T8 between the TD group and the ASD group. Channel T8 was selected because relatively high correlation coefficient values were observed in both the ASD group and the TD group in this channel. The results of the TD participants provided a baseline for how the frequency and sound intensity hearing levels would affect the peak amplitude and peak latency of the component. Generally, the higher the sound intensity hearing level, the higher the absolute peak amplitude of the component. In addition, the higher the sound intensity hearing level, the lower the peak latency. This was clearly observed in the N1 component. When looking at the N1 peak amplitude, the TD group had an overall smaller amplitude compared to the ASD group.

[0066] Autistic participants may have different sound responses and individual sound sensitivities to different types of sound sources. Some participants may like a certain type of sound, while others may find it unpleasant. This indicates that the physical characteristics of the sound that trigger problem behaviors vary from person to person. Therefore, a headset with the same noise control strategy and algorithm is not suitable for participants with different auditory perception responses and sensations. Therefore, it is crucial to provide customized noise control for autistic participants with different frequency distributions. To achieve this, clustering analysis was performed on autistic clustered participants divided into different subgroups according to the auditory perception of autistic participants, and each group would have a similar frequency distribution.

[0067] Clustering algorithms include prototype-based clustering and hierarchical clustering, which differ in the properties of the grouping mechanism. One of the methods of prototype-based clustering is K-means clustering, which can be used as a partitioning algorithm. It is a vector quantization method used to divide a certain number of participants into K clusters, where each participant belongs to the cluster with the closest mean or centroid. This method requires the number of clusters (K), cluster initialization, and distance metric as input parameters. Let X = {x i}, i = 1, …, n be the dataset that needs to be clustered into a group of K clusters, where x i is the vector of the average auditory perception scores of the i-th ASD participant, and n is the total number of ASD participants used in the analysis. Let C = {c k} be the cluster set, k = 1, …, K be the number of clusters to be formed, and μ k be the mean of cluster c k . The squared error between μ k and the points in cluster c k is defined as

[0068]

[0069] The goal of k-means is to minimize the sum of the squared errors (SSE) of all K clusters,

[0070]

[0071] and find a partition that minimizes the sum of squared errors between the empirical mean of the clusters and the data points within the clusters. After determining the parameter K, the K-means algorithm first initializes K average auditory perception score vectors randomly selected from the dataset as the initial cluster centers (μ k ). For each ASD participant, the Euclidean distance between the average auditory perception score vector and all cluster centers is calculated. The ASD participant is assigned to the cluster with the minimum Euclidean distance. After all ASD participants are assigned to clusters, the cluster centers are recalculated using the current cluster membership. Then, the process of calculating the Euclidean distance between ASD participants and cluster centers is repeated until the cluster assignments of all ASD participants do not change. The block diagram of the K-means clustering algorithm is shown in Figure 6 .

[0072] Among the input parameters, the most crucial parameter is the number of clusters K. Currently, there is no perfect mathematical criterion to determine K. A typical heuristic for choosing K is to run the algorithm independently for different values of K and select the partition that seems to be the most meaningful solution to the problem. This method was adopted in this study, and the method used to select K was based on the cluster validation index.

[0073] Another input is the cluster initialization. Since K-means only converges to a local minimum, different initializations may lead to different clustering solutions. To overcome this problem, each number of clusters is initialized with 10,000 different initial centroid positions. This number of initializations was chosen because it can provide stable clustering solutions and membership assignments in the current cluster analysis for different values of K. Subsequently, the partition with the minimum sum of squared errors is selected.

[0074] Another way to analyze the current data is to use the Agglomerative Hierarchical Algorithm (HCA). This analysis involves constructing a clustering hierarchy using a "bottom-up" approach. It starts with each data point as a separate cluster and merges them into progressively larger clusters until all the data is grouped into one large cluster. At each clustering step, the clusters with the smallest distance are joined together, and there are several ways to determine the distance between two clusters, which are called linkages. Several indices can be used to examine or determine how to merge or split clusters. For example, average linkage measures the cluster distance as the average of all pairwise distances between the data points in the two clusters, and Ward linkage, which is based on the Euclidean distance between the centroids of the two clusters multiplied by a factor. The closest pair of clusters calculated using this method results in the smallest increase in the total SSE of the dataset. Comparing these linkage methods, Ward linkage and average linkage usually capture the clustering structure more effectively than single linkage and complete linkage. Therefore, average linkage and Ward linkage are used in the Agglomerative Hierarchical Clustering algorithm. This study uses the Silhouette index, Calinski-Harabasz index, and Davies-Bouldin index to select the appropriate clustering algorithm and the optimal number of clusters because they have been shown to be some of the best-performing clustering validation indices in both artificial and real-world datasets. These three indices provide better results even in datasets with typically problematic features such as high dimensionality, density asymmetry, and cluster overlap, which may also be present in our dataset.

[0075] The comparison of the clustering results of the three methods is shown in Table 2 below, which shows the cluster membership assignments for the selected K. The clustering methods are distributed by row, while the individual clusters are distributed by column. Among the three clustering methods, only the HCA average linkage tends to form large clusters that include most of the participants, as well as small clusters that contain 1 to 4 members regardless of the value of K. Since the purpose of performing the clustering is to group participants with similar frequency distributions, the clustering solution should not contain a single cluster that includes almost all participants - especially those with heterogeneous auditory perception responses. Therefore, this solution is not suitable for the current purpose. For the other two algorithms, the partitions have similar sizes. The results of the clustering validation indices are as Figure 7 shown. In the case of the Silhouette and Calinski-Harabasz indices, higher values indicate a better partition, while in the case of the Davies-Bouldin index, lower values indicate a better partition. As indicated by the clustering validation indices, the K-means algorithm presents a slightly better performance than the HCA-Ward linkage, so the clustering solution of the former is selected. For the number of clusters to be formed, two of the three clustering validation indices suggest that K = 5 is the optimal cluster assignment. Therefore, based on the results of the K-means clustering algorithm, the ASD group is divided into five clusters.

[0076] Table 2 Cluster member assignments from clustering algorithms: K-means, HCA-average linkage, and HCA-Ward linkage; K = 3 to 6

[0077]

[0078] Based on the K-means clustering method, the characteristics of the clustering groups of autistic participants with corresponding frequency distributions at different dB HL were investigated. Figure 8 It was shown that the first cluster (ASD C1) rated the frequencies of 4 kHz and 8 kHz as unpleasant at low sound intensity levels, while rating the frequencies of 250 Hz, 500 Hz, and 2 kHz as neutral and 1 kHz as liked. At higher sound intensity levels, all frequencies were rated as unpleasant, with 250 Hz, 4 kHz, and 8 kHz being the most unpleasant. The second cluster (ASD C2) rated the frequencies as neutral or similar at low sound intensity levels, rated 250 Hz and 8 kHz as unpleasant at 60 dB HL, rated 2 kHz and 4 kHz as unpleasant at 70 dB HL, and rated all frequencies as disliked at the highest sound intensity level. The third cluster (ASD C3) rated all sounds above 2 points (disliked), except for 8 kHz at certain sound intensity levels. The fourth cluster (ASD C4) rated the frequencies as neutral or liked at low sound intensity levels, rated 250 Hz, 500 Hz, and 2 kHz as unpleasant at 60 dB HL, and rated all frequencies as disliked at higher sound intensity levels. The fifth cluster (ASD C5) rated most frequencies as liked, with 250 Hz being the only frequency rated as disliked at higher sound intensity levels.

[0079] For all clusters, the frequencies that elicited the most unpleasant sensations at higher sound intensity hearing levels were 250 Hz and 8 kHz, followed by the 2 kHz frequency, which was the second most unpleasant frequency. At other sound intensity hearing levels, the auditory perception responses from different groups had their own characteristics. This supports the need for customized noise control that addresses specific annoying frequencies for each group. Additionally, the amount of noise reduction that can result in a neutral rating for the perceived sound of the presented frequencies varies by group, with some groups requiring a greater reduction and some groups preferring a moderate reduction. Therefore, it is crucial to consider these different responses when designing noise control methods for autistic participants.

[0080] To provide noise control for the heterogeneous needs of autistic participants, the hearing perception curves of each corresponding subgroup were plotted based on the above findings. Since the ASD group has an auditory perception response curve different from that of the TD group, the noise control strategy focuses on providing a suitable noise control algorithm to eliminate the incoming noise so that ASD participants have a neutral response to the generated sound. Figure 9A Shows the relationship between the auditory perception level and dB HL at different frequencies. To achieve a neutral response (e.g., Figure 9A the dashed line in as the standard, the level of noise reduction required for each frequency was studied. In this regard, a power function curve fitting was performed with the average auditory perception level and the sound intensity hearing level as variables. The fitted curve was used to estimate the average auditory perception level between the sound intensity hearing levels tested at each frequency in the auditory perception test. The form of the power function is i where y i is the i-th average perception level at a specific frequency, x Figure 9A is the i-th intensity level at a specific frequency, and a, b, c are coefficients to be determined. Using the nonlinear least squares method, the best fit curve was determined by selecting the best fit with the minimum sum of squared errors. The power function was chosen because the auditory perception level is inversely proportional to the presented sound intensity hearing level. As the sound intensity hearing level decreases, the change in the auditory perception level generally decreases. Using the hearing perception curve, the noise attenuation required to cause a neutral sensation at individual frequencies can be estimated given the noise level presented to the participant. As Figure 9A and 9B show the hearing perception curve and the auditory perception level and the resulting noise cancellation target curve.

[0081] Participants with autism and auditory over-responsivity typically use noise-canceling headphones to reduce their exposure to noise and its negative effects. Commercially available noise-canceling headphones allow users to change the overall noise-canceling function by adjusting the level of noise cancellation applied by the headphones. This allows users to vary the reduction in the overall sound pressure level. So far, no noise control strategy has been designed based on the human perception response curve, which is a function of sound amplitude and frequency. The present inventors have demonstrated that frequency is also a factor that significantly affects the auditory perception of autistic participants with auditory over-responsivity. Therefore, in addition to intensity, the ability to adjust the frequency response of the noise-canceling function would be more beneficial for achieving the objective. To develop an ANC algorithm to mitigate the aversive behaviors associated with auditory over-responsivity in autistic participants, the frequency response and noise cancellation level are adjusted based on the results of auditory perception tests. The goal of the ANC algorithm is to achieve noise cancellation such that autistic participants can perceive the incoming noise with a neutral sensation, thereby minimizing the impact of the incoming interfering noise on their behavior. Figure 10A and 10B Figure 1 shows a block diagram of an ANC system with the proposed functionality. The sound signal includes the incoming noise. The sound signal can be regarded as the main signal X(z) and is measured using the reference microphone 1022 of the headphones 1020. This sound signal propagates from the outside of the earcup to its inside along the main path P(z) ( Figure 10B represented by 1002 in Figure 1). The residual signal D(z) represents the signal remaining after a portion of the sound signal is absorbed by the ear pads. The computer device 1040 outputs the cancellation signal Y(z), which travels through the secondary path (S(z)) ( Figure 10B shown as 1004 in Figure 1), which includes electronic software and hardware components, as well as the sound path from the speaker to the error microphone 1024. Y'(z) represents the cancellation signal after passing through the secondary path S(z). It combines with the residual signal D(z) and generates the cancellation of the noise based on the superposition principle, thereby generating the error signal E(z). The adaptive noise control system can include feedback, feedforward, or hybrid control. In feedback control, since no feedback and reference signals are required, the error microphone signal is used to generate the cancellation signal. However, its active attenuation performance is limited by the resonance behavior of the earcup cavity, which forces the feedback gain to be low. Generally, if a good reference signal is available and the system is efficient in meeting causality, feedforward control can produce better performance than feedback control. However, it may suffer from stability or performance defects due to the limited tolerance to gain errors. The hybrid control system retains the advantages of the feedforward and feedback control systems, overcomes the instability of the feedback system, and compensates for the poor adaptability of the feedforward system to the original noise. Therefore, a hybrid system is adopted in this study. The adaptive filter W(z) (in Figure 10BThe purpose (represented by 1030 in

[0082] It will be understood that in Figure 10B , the reference microphone 1022 and the error microphone 1024 can be part of the headset 1020. Alternatively, one or both of them can be part of the computer device 1040. Additionally, the computer device 1040 and the headset 1020 can be integrated such that the computer device 1040 forms part of the headset 1020.

[0083] Verification tests are conducted to check the noise cancellation adjustment performance based on the auditory perception curve. The sound presentation system, experimental environment, and procedures are similar to those of the auditory perception test described above, except that the presented auditory stimuli are processed for noise attenuation according to the target curve. Some of the participants recruited for this study are invited to participate in the verification. Figure 11The comparison of the auditory perception responses of ASD participants to the original and processed sound stimuli is shown. The results indicate that the proposed adjustable sound perception method is particularly effective for medium-high noise levels. The aim of the target noise reduction level is to provide a neutral sensation or response, rather than achieving the maximum noise reduction in the current project. It has been found that at the frequency and sound intensity hearing levels tested using the proposed noise control method, the perceived score grades are still close to the target of a neutral sensation (score of 3). A slightly larger deviation was observed at the 250 Hz frequency compared to other frequencies at 30 and 78 dB HL; however, an improvement was still observed in the auditory perception response after applying the noise control. Similarly, better auditory perception responses were observed at all frequencies in the 60 - 78 dB HL range except at the 1 kHz frequency. This may be related to the effective frequency ranges of the active and passive noise control methods. Active noise cancellation is more effective in the low-frequency range, such as at the frequencies of 250 and 500 Hz. When the noise frequency approaches 1 kHz, the noise cancellation level becomes increasingly limited. Passive noise control can provide significant noise attenuation in the high-frequency range from 2000 to 8000 Hz. The proposed adjustable sound perception method is effective, which allows for the reduction of noise at specific frequencies based on the characteristic auditory perception of autistic participants. Thus, with this attenuation of the auditory stimulus, autistic participants will experience a comfortable auditory environment, which can alleviate their aversive behaviors.

[0084] As used herein, the terms "sound", "auditory", "auditorial", etc. may be used interchangeably.

[0085] It will also be understood that any features in the above embodiments of the present disclosure can be combined together and need not be applied in isolation from each other. Those skilled in the art can easily make similar combinations of two or more features of the above embodiments or preferred forms of the present disclosure.

[0086] Unless otherwise defined, the technical and scientific terms used herein have the ordinary meanings commonly understood by those skilled in the art related to the exemplary embodiments. The embodiments are illustrated by way of non-limiting examples. Based on the above-disclosed embodiments, various modifications that can be conceived by those skilled in the art fall within the spirit of the exemplary embodiments.

Claims

1. A method for controlling noise for a human subject, comprising: The human subject receives a sound signal through a headset, the sound signal including a noise signal; Generating a cancellation signal based on a hearing target curve related to sound amplitude and frequency; And Applying the cancellation signal to the sound signal such that the noise signal is attenuated.

2. The method according to claim 1, wherein The hearing target curve includes the relationship between noise attenuation in decibels (dB) and sound frequency.

3. The method according to claim 2, wherein The sound frequency ranges from 250 Hz to 8000 Hz.

4. The method according to claim 2, further comprising: Determining a sound intensity hearing level corresponding to the neutral response of the human subject based on an auditory perception curve; And Calculating the noise attenuation based on the difference between the sound intensity hearing level and the noise level.

5. The method according to claim 4, further comprising: By performing power function curve fitting to determine the neutral response, where y i is the i-th average perceived level, x i is the i-th intensity level, and a, b, and c are coefficients.

6. The method according to claim 4, further comprising performing a clustering algorithm on multiple human subjects.

7. The method according to claim 6, wherein, The clustering algorithm includes an agglomerative hierarchical algorithm.

8. The method according to claim 6, further comprising performing an electroencephalogram test on the multiple human subjects to obtain the neural responses of the multiple human subjects in response to sound stimuli.

9. The method according to claim 8, further comprising: Recording the data measured by the electroencephalogram test; Re-referencing the data to obtain re-referenced data; Filtering the re-referenced data to obtain filtered data; And Determining a search window to identify a first peak P1, a second peak P2, and a trough N1.

10. The method according to claim 9, further comprising performing baseline correction on the filtered data.

11. The method according to claim 10, further comprising averaging the filtered data in the search window.

12. The method according to claim 1, further comprising: Configuring the sound signal such that the sound signal propagates along a main path and evolves into a residual signal; And Performing superposition of the residual signal and the cancellation signal to generate an error signal.

13. The method according to claim 12, further comprising: Measuring the error signal; Calculating the difference between the sound signal and the error signal; And Comparing the difference and the cancellation signal.

14. The method according to claim 13, further comprising: If the difference between the difference and the cancellation signal is greater than a threshold, adjusting at least one parameter of an adaptive filter.

15. A system for controlling noise for a human subject, comprising: A headset configured to receive a sound signal; And A computer device configured to generate a cancellation signal based on a hearing target curve related to sound amplitude and frequency and apply the cancellation signal to the sound signal such that the noise signal is attenuated.

16. The system according to claim 15, further comprising: A main path configured to transmit the sound signal such that the sound signal evolves into a residual signal; And A secondary path configured to transmit the cancellation signal, Among them, at the intersection of the main path and the secondary path, the residual signal and the cancellation signal are superimposed to generate an error signal.

17. The system according to claim 16, wherein, The headset includes: a reference microphone configured to measure the sound signal; and an error microphone configured to measure the error signal.

18. The system according to claim 17, further comprising an adaptive filter provided on the secondary path.

19. The system according to claim 18, wherein, The computer device is configured to calculate the difference between the sound signal and the error signal; and compare the difference with the cancellation signal.

20. The system according to claim 19, wherein The computer device is configured to adjust at least one parameter of the adaptive filter when the difference and the cancellation signal are greater than a threshold.

Images