A source separation method based on adaptive energy filtering of swept-frequency SFOAEs

Abstract

The application provides a source separation method based on adaptive energy filtering of sweep frequency SFOAEs, which can accurately separate SFOAEs components of different cochlea generation sites, and provides a data basis for subsequent accurate prediction. In the technical scheme of the application, sweep frequency stimulating sound S1 and sweep frequency suppressing sound S2 which change continuously and linearly with time are combined into a stimulating signal in the time domain; the stimulating sound is played in a four-segment dual-tone mode, the linear stimulating sound is quickly and accurately eliminated based on the four-segment dual-tone suppression paradigm, and the SFOAEs induced by the suppressing sound are preliminarily offset, so that the residual signal of the SFOAEs induced by the stimulating sound is extracted; the SFOAEs signal induced by the sweep frequency stimulating sound is separated and extracted through a dynamic tracking filter, so that the target SFOAEs signal is obtained; the target SFOAEs signal is subjected to short-time Fourier transform, so that time domain and frequency domain information are obtained at the same time, the amplitude spectrum and the phase spectrum of the sweep frequency SFOAEs signal are calculated according to the time-frequency distribution diagram, and the SFOAEs components of different positions are accurately separated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal processing, in particular to a source separation method based on adaptive energy filtering of sweep SFOAEs. BACKGROUND

[0002] Stimuli frequency otoacoustic emissions (SFOAEs) can characterize the change of absolute hearing threshold in cochlear hearing loss (CHL). In the existing SFOAEs research, different frequency pure tones are input one by one as stimulus sound in the CHL prediction process, played to the subjects, and then the response signals of the cochlea in the ear canal of the subjects are collected one by one, that is, SFOAEs, and the characteristics of the response signals are analyzed and extracted to determine the hearing damage of the subjects. However, because in the prior art, SFOAEs signals are induced based on single frequency pure tone stimulus sound, the vector sum of the response signals of all sources generated by the cochlea, that is, there is mutual interference between them, so the loss of signal characteristics may occur between these source mixed signals, thereby causing the CHL to be unable to be accurately predicted. SUMMARY

[0003] In order to solve the problem that the SFOAEs detection method in the prior art is limited to single variable analysis and linear prediction algorithm, and the SFOAEs signals generated by the mutual separation of the sources cannot be collected, the present application provides a source separation method based on adaptive energy filtering of sweep SFOAEs, which can accurately separate the SFOAEs components of different cochlea generation sites, and provide data basis for subsequent accurate prediction.

[0004] The technical scheme of the present application is as follows: a source separation method based on adaptive energy filtering of sweep SFOAEs, characterized in that it comprises the following steps:

[0005] S1: constructing a sweep tone with continuous and linear change of frequency over time as a stimulus signal;

[0006] Two linear sweep signals include: sweep stimulus sound S1 and sweep suppression sound S2, the frequency f p increases linearly from f1 to f2 in a time period T, the frequency f s is always higher than f p by NHz, where N is a positive number;

[0007] In the stimulus signal, the sweep stimulus sound S1 and the sweep suppression sound S2 are synthesized in the time domain as:

[0008]

[0009] Where p is the synthesized stimulus signal, t is the time variable,

[0010] O(t) is a window function multiplied on the swept signal, whose beginning and ending are a half cosine window, P0 is the sound pressure amplitude,

[0011] is the instantaneous phase:

[0012]

[0013] b is the scanning rate, is the adjustable initial phase;

[0014] S2: using a four-segment two-tone suppression paradigm to eliminate the interference of linear stimulus sound, preliminarily canceling the SFOAEs induced by the suppression sound, extracting the residual signal of the SFOAEs induced by the stimulus sound, and obtaining the target SFOAEs signal;

[0015] Specifically comprising the following steps:

[0016] a1: using two earphones a and b to respectively play swept stimulus sound S1 and swept suppression sound S2;

[0017] Each single stimulus signal playing process includes four consecutive stages: A segment, B segment, C segment and D segment.

[0018] In the A segment and the B segment, only the earphone a is used to play S1, and the S1 in the A segment and the B segment is reversed, and the earphone b does not play S2.

[0019] In the C segment and the D segment, the earphone a and the earphone b simultaneously play S1 and S2 respectively, wherein the S1 in the C segment and the D segment is reversed, and the S2 is in phase.

[0020] a2: in each said stimulus signal playing process, a miniature low-noise microphone is used to collect the response of the cochlea in the ear canal in the A, B, C and D segments, and is recorded as: p A (t), p B (t), p C (t), p D (t)

[0021] p A contains background noise, and contains the positive phase of the stimulus sound artifact and the SFOAEs signal SFE + induced thereby;

[0022] p B contains background noise, and contains the negative phase of the stimulus sound artifact and the negative phase SFOAEs signal SFE - induced thereby;

[0023] p C mainly contains the stimulus sound artifact Inhibition artifact and S2-induced SFOAEs signal SFEs +

[0024] p D mainly including anti-phase stimulation artifact positive phase inhibition artifact and S2-induced SFOAEs signal SFEs +

[0025] a3: the corresponding signal in the ear canal collected by the miniature low-noise microphone is passed through a band-pass filter with a bandwidth of f1-f2 Hz to eliminate high-frequency and low-frequency background noise interference; wherein f2>f1;

[0026] a4: calculate the time-domain sweep SFOAEs residual signal 2p SFOAE (t), the specific calculation method is as follows:

[0027]

[0028] a5: using a dynamic tracking filter, the obtained residual signal 2p SFOAE (t) is dynamically tracked and filtered to remove the SFOAEs signal of sweep inhibition sound S2 that has not been completely canceled, and the target SFOAEs signal is obtained;

[0029] S3: processing the target SFOAEs signal, simultaneously obtaining time domain and frequency domain information, and calculating the amplitude spectrum and phase spectrum of the sweep SFOAEs signal according to the time-frequency distribution diagram;

[0030] Specifically, the following steps are included:

[0031] b1: using IFFT to transform the target SFOAEs signal of complex spectrum from frequency domain to time domain or latency domain, and obtaining SFOAEs time-frequency distribution diagram based on CWT;

[0032] b2: based on least square fitting, the SFOAEs time-frequency distribution diagram is obtained to obtain a filter cutoff line suitable for all stimulation intensities;

[0033] b3: based on the difference of ear canal position, different extraction targets are set, based on the fitted filter f-t cutoff line function, the wavelet coefficient of the latency component that needs to be filtered out is set to zero on the time-frequency distribution diagram, and the target wavelet coefficient component is obtained;

[0034] b4: the target wavelet coefficient component is calculated through inverse CWT to obtain the time domain signal of SOFEs component;

[0035] ​​b5: the time-domain signal of the SOFEs component is processed by FFT operation to obtain the amplitude spectrum and phase spectrum of the SFOAEs component generated at different positions and with different latencies.

[0036] It is further characterized in that:

[0037] The calculation method of the scanning rate b is as follows:

[0038]

[0039] Wherein, f1=f(t1)

[0040] f(t)=f1+b(t-t1)

[0041] f(t) is the instantaneous frequency of the stimulation signal, which increases linearly with time, t1 and t2 are the starting and ending times of the scanning period;

[0042] In step a1, when each single stimulation signal is played:

[0043] A silent section with a length of 2T d is added before the start of section A, and a silent section with a length of T d is added after the end of section D, wherein T d is the system delay measured in advance;

[0044] In step a5, the dynamic tracking filter uses a real-time dynamic tracking filter with a continuously changing center frequency over time;

[0045] The fitting filter f-t cutoff line function is:

[0046] τ i =c i ×a×f b

[0047] f is the frequency of SFOAEs, τ is the latency of SFOAEs, i is the serial number of the filter cutoff line, and a, b, and c are parameters to be determined for the fitting function.

[0048] The application provides a source separation method based on adaptive energy filtering of sweep SFOAEs, which uses sweep stimulation sound S1 and sweep suppression sound S2 with continuous and linear change of frequency over time to be synthesized into a stimulation signal in the time domain, and one-time playing can provide sufficient frequency of the stimulation signal for detection, ensures that the data amount included in the SFOAEs residual signal induced by the stimulation sound is sufficient for subsequent position separation, and compared with the prior art using a single-frequency signal as the stimulation sound playing mode, the application uses the sweep signal as the stimulation sound, and greatly improves the detection efficiency; in the application, the stimulation sound playing mode is designed, the stimulation sound is played in a four-segment dual-tone mode, the interference of the linear stimulation sound is quickly and accurately eliminated based on the four-segment dual-tone suppression paradigm, the SFOAEs induced by the suppression sound are preliminarily offset, the SFOAEs residual signal induced by the stimulation sound is extracted, the whole process is high in efficiency and simple in operation; the sweep stimulation sound induced SFOAEs signal is separated and extracted through a dynamic tracking filter, the target SFOAEs signal is obtained, and the target SFOAEs signal is ensured to be clean and accurate; finally, the short-time Fourier transform is performed on the target SFOAEs signal, the time domain and frequency domain information can be obtained at the same time, the amplitude spectrum and phase spectrum of the sweep SFOAEs signal are calculated according to the time-frequency distribution diagram, and different position SFOAEs components are accurately separated, thereby providing a data basis for subsequent accurate prediction. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 It is a fast detection flow chart of the target SFOAEs signal;

[0050] Figure 2 It is a change relationship of the frequency fp of the linear sweep stimulation sound S1 and the frequency fs of the suppression sound S2 with respect to time;

[0051] Figure 3 It is a stimulation sound playing timing diagram of the four-segment sweep SFOAEs test algorithm based on dual-tone suppression;

[0052] Figure 4 a is a time-frequency distribution diagram of the SFOAEs residual signal after time domain preprocessing;

[0053] Figure 4 b is an extraction schematic diagram of the sweep SFOAEs by the dynamic tracking filter;

[0054] Figure 5 It is a flow chart of the SFOAEs source separation technology. DETAILED DESCRIPTION

[0055] The application includes a source separation method based on adaptive energy filtering of sweep SFOAEs, which mainly includes Figure 1 the fast detection process of the target SFOAEs signal andFigure 5 The SFOAEs source separation process is shown.

[0056] S1: Construct a sweep tone whose frequency changes continuously and linearly over time as a stimulus signal;

[0057] like Figure 2 As shown, the two linear frequency sweep signals include: a probe sweep S1 and a suppressor sweep S2, with the frequency f of the probe sweep S1 being... p The frequency f of the sweep suppression sound S2 increases linearly from f1 to f2 over a time period of duration T. s Then it is always more than f p It is higher than N Hz, where N is a positive number. In this embodiment, N is 47Hz.

[0058] In the stimulus signal, the swept-frequency stimulus sound S1 and the swept-frequency suppression sound S2 are combined in the time domain as follows:

[0059]

[0060] Where p is the synthesized stimulus signal; t is the time variable;

[0061] O(t) is a window function multiplied on the sweep signal. Its start and end points are half-cosine windows, which make the rising and falling edges of the sweep signal change slowly, thereby avoiding interference introduced by switching noise.

[0062] P0 is the sound pressure level; It is the instantaneous phase:

[0063]

[0064] In the formula It is an adjustable initial phase;

[0065] b is the scan rate, in Hz / s; the scan rate b is calculated as follows:

[0066]

[0067] Where, f1 = f(t1),

[0068] f(t) = f1 + b(t - t1)

[0069] The stimulus sound in this application is a swept frequency signal with a variable frequency. Therefore, f(t) is used to represent the instantaneous frequency of the stimulus signal, which increases linearly with time. t1 and t2 are the start and end times of the scanning period.

[0070] Based on the calculation formula of the scanning rate b, it can be known that the scanning rate in a fixed frequency range depends on the scanning time T (T = t2-t1).

[0071] S2: using a four-section two-tone suppression paradigm to eliminate the interference of linear stimulus sound, preliminarily offsetting the SFOAEs induced by the suppression sound, extracting the residual signal of the SFOAEs induced by the stimulus sound, and obtaining the target SFOAEs signal;

[0072] Specifically, the following steps are included:

[0073] a1: using two earphones a and b to respectively play the swept-frequency stimulus sound S1 and the swept-frequency suppression sound S2;

[0074] Each single stimulus signal playing process includes four consecutive stages: A section, B section, C section, and D section.

[0075] In the A section and the B section, only the earphone a is used to play S1, and S1 in the A section and the B section is reversed, and the earphone b does not play S2.

[0076] In the C section and the D section, the earphones a and b simultaneously play S1 and S2 respectively, wherein S1 in the C section and the D section is reversed, and S2 is in phase.

[0077] A mute section with a time length of 2T d is added before the start of the A section, and a mute section with a time length of T d is added after the end of the D section, wherein T d is a system delay measured in advance; by adding a mute section at the beginning and the end of the single stimulus signal playing process respectively, the influence of the system delay is eliminated, and the extracted signal is more accurate.

[0078] Specifically, each single stimulus signal playing process is shown in the attached Figure 3 The stimulus sound playing timing diagram of the four-section swept-frequency SFOAEs test algorithm based on two-tone suppression is shown, wherein the earphone a is used to play the swept-frequency stimulus sound S1, and the earphone b is used to play the swept-frequency suppression sound S2.

[0079] a2: in each stimulus signal playing process, a miniature low-noise microphone is used to collect the responses of the cochlea in the ear canal in the A, B, C, and D sections, which are sequentially recorded as: p A (t), p B (t), p C (t), and p D (t).

[0080] p A contains the positive-phase stimulus sound artifact and the SFOAEs signal SFE + induced thereby in addition to the background noise;

[0081] p B In the middle, in addition to containing background noise, also contains the reversed stimulation sound artifact and the induced reversed SFOAEs signal SFE - .

[0082] In C and D segments, sweep stimulation sound S1 and sweep suppression sound S2 are played simultaneously. According to the principle of two-tone suppression of cochlea, when the amplitude of sweep suppression sound S2 is large enough, S2 can completely suppress the SFOAEs signal SFE generated by S1, and the effect of S1 on S2 can be ignored, that is, the SFOAEs signal SFE induced by S1 in C and D segments is almost completely suppressed, and the SFOAEs signal SFEs induced by S2 remains basically unchanged.

[0083] p C In the middle, mainly contains stimulation sound artifact suppression sound artifact and S2 induced SFOAEs signal SFEs + ;

[0084] p D In the middle, mainly includes reversed stimulation sound artifact positive suppression sound artifact and S2 induced SFOAEs signal SFEs + .

[0085] That is, the technical scheme of the present application realizes the rapid detection of sweep tone SFOAEs based on the two-tone suppression algorithm through the design of the stimulation signal playing process, greatly improving the detection efficiency.

[0086] a3: The corresponding signal collected in the ear canal by the miniature low-noise microphone is passed through a band-pass filter with a bandwidth of f1-f2 Hz to eliminate high-frequency and low-frequency background noise interference; wherein f2>f1.

[0087] a4: In the present application, the sweep SFOAEs residual signal 2p SFOAE (t) in time domain is calculated by using four-section two-tone suppression paradigm, and the specific calculation method is as follows:

[0088]

[0089] In the present application, through the accurate setting of the playing mode of the stimulation sound, based on the four-section two-tone suppression paradigm, the stimulation sound artifact suppression sound artifact and the SFEs induced by the suppression sound S2 are all eliminated, and only the SFOAEs signal SFE induced by the stimulation sound S1 is left. The four-section two-tone suppression paradigm effectively eliminates the interference of the second harmonic. The four-section two-tone suppression paradigm design is also called linear cancellation.

[0090] a5: using a dynamic tracking filter to obtain the residual signal 2p SFOAE (t) performing dynamic tracking filter processing to remove the SFOAEs signal of the sweep tone S2 which is not completely cancelled, to obtain the target SFOAEs signal.

[0091] The short-time Fourier transform is performed on the SFOAEs residual signal after time domain preprocessing to obtain a time-frequency distribution diagram as shown in Figure 4 (a). After linear cancellation operation based on four-section two-tone suppression, the SFOAEs signal SFEs induced by the suppression tone S2 is theoretically completely cancelled, but in practice there may be a small amount of SFEs which is not completely cancelled. To further eliminate the SFOAEs signal induced by S2, the dynamic tracking filter is used to extract the target SFOAEs signal. The time-frequency distribution diagram of Figure 4 (a) is simplified as Figure 4 (b), SFE represents the SFOAEs induced by the stimulation tone, i.e. the target SFOAEs signal to be extracted, and SFEs represents a small amount of SFOAEs induced by the suppression tone.

[0092] In a specific embodiment, the dynamic tracking filter includes a dynamic zero point and a dynamic pole; the dynamic zero point corresponds to a band-stop filter, and the zero point frequency automatically follows the frequency of the filtered signal; the dynamic pole corresponds to a band-pass filter, and the pole frequency automatically follows the frequency of the signal to be extracted; in a specific implementation, the bandwidth of the band-pass filter is set, and the zero and pole of the filter are initialized, and then subsequent filtering operation is performed.

[0093] S3: processing the target SFOAEs signal to obtain time domain and frequency domain information, and calculating the amplitude spectrum and phase spectrum of the sweep SFOAEs signal according to the time-frequency distribution diagram.

[0094] In this application, based on continuous wavelet transform, an adaptive energy filtering SFOAEs source separation technology is proposed. The sweep SFOAEs spectrum in a certain frequency range is a function of frequency, which is converted into an equivalent SFOAEs waveform which is called time domain (or latency domain) SFOAEs by IFFT.

[0095] The SFOAEs source separation technology is shown in Figure 5 , which specifically includes the following steps:

[0096] b1: using IFFT to convert the target SFOAEs signal of the complex frequency spectrum from the frequency domain to the time domain or latency domain, and obtaining the SFOAEs time-frequency distribution diagram based on CWT;

[0097] The source separation process of SFOAEs, i.e. separating and extracting the latency components generated at different locations, is mainly achieved by setting the wavelet coefficients of the latency components to be filtered out to zero on the time-frequency distribution diagram, so the key point is how to fit a series of suitable filter cutoff lines on the time-frequency distribution diagram.

[0098] b2: based on the least square fitting of the SFOAEs time-frequency distribution diagram, a filter cutoff line suitable for all stimulation intensities is obtained;

[0099] The fitted filter f-τ cutoff line function is: τ i = c i ×a×f b

[0100] f is the frequency of SFOAEs, τ is the latency of SFOAEs, i is the serial number of the filter cutoff line, and a, b, c are parameters to be determined in the fitting function

[0101] For example, in order to extract the main component of SFOAEs, the goal of fitting is to determine two filter cutoff lines τ0 and τ1 by traversing the preset parameter combinations of a, b and c, so that the energy contained in the wavelet coefficients between the two lines is maximized.

[0102] b3: based on the different ear canal locations, different extraction targets are set, based on the fitted filter f-τ cutoff line function, the wavelet coefficients of the latency components to be filtered out are set to zero on the time-frequency distribution diagram, and the target wavelet coefficient component is obtained;

[0103] b4: the target wavelet coefficient component is calculated through inverse CWT to obtain the time domain signal of the SOFEs component;

[0104] b5: the time domain signal of the SOFEs component is processed through FFT operation to obtain the amplitude spectrum and phase spectrum of the SFOAEs components generated at different locations and having different latencies.

[0105] The SFOAEs source separation technology proposed in the present application is suitable for SFOAEs signals of all stimulation intensities, which uses continuous wavelet transform to obtain the SFOAEs time-frequency distribution diagram, and uses the empirical relationship between latency and frequency as a basis to fit a filter cutoff line suitable for all stimulation intensities on the time-frequency distribution diagram. Finally, the source separation technology is used to separate the SFOAEs components at different locations from the total SFOAEs.

[0106] Using the technical solution of the application, a sweep tone with continuous and linearly changing frequency is used as a stimulation signal, a four-section dual-tone suppression paradigm is used to eliminate the interference of linear stimulation sound, and the SFOAEs induced by the suppression sound are preliminarily offset to extract the residual signal of the SFOAEs induced by the stimulation sound. Then, a real-time dynamic tracking filter with a continuously changing center frequency over time is used to separate and extract the sweep suppression sound-induced SFOAEs signal to obtain the target SFOAEs signal. Finally, the target SFOAEs signal is subjected to short-time Fourier transform to obtain time domain and frequency domain information at the same time, and the amplitude spectrum and phase spectrum of the sweep SFOAEs signal are calculated according to the time-frequency distribution diagram to prepare for accurate prediction of the CHL. The source separation method based on sweep SFOAEs adaptive energy filtering proposed in the application is easy to understand, can accurately separate the SFOAEs components at different positions from the total SFOAEs, the obtained result is intuitive and obvious, and the execution efficiency is high, thereby providing accurate data basis for subsequent CHL prediction.

Claims

1. A source separation method for adaptive energy filtering based on swept-frequency SFOAEs, characterized in that, It includes the following steps: S1: Construct a sweep tone whose frequency changes continuously and linearly over time as a stimulus signal; The two linear frequency sweep signals include: a frequency sweep stimulus sound S1 and a frequency sweep suppression sound S2, wherein the frequency of the frequency sweep stimulus sound S1 is f. p The frequency f of the sweep suppression sound S2 increases linearly from f1 to f2 over a time period of duration T. s Then it is always more than f p It is N Hz higher than that, where N is a positive number; In the stimulation signal, the frequency sweep stimulation sound S1 and the frequency sweep suppression sound S2 are combined in the time domain to form: Where p is the synthesized stimulus signal, and t is the time variable. O(t) is a window function multiplied on the swept frequency signal, with a half-cosine window at the beginning and end, and P0 is the sound pressure level. It is the instantaneous phase: b is the scan rate. It is an adjustable initial phase; t1 is the start time of the scan period; S2: Use the four-segment dual-tone suppression paradigm to eliminate the interference of linear stimulus sound, initially cancel the SFOAEs induced by the suppression sound, extract the residual signal of the SFOAEs induced by the stimulus sound, and obtain the target SFOAEs signal. Specifically, the following steps are included: a1: Use two headphones, a and b, to play the sweep frequency stimulation sound S1 and the sweep frequency suppression sound S2 respectively; Each single stimulus signal playback process consists of four consecutive phases: phase A, phase B, phase C, and phase D; In segments A and B, only earphone a plays S1, and S1 in segments A and B is out of phase; earphone b does not play S2. In segments C and D, headphones a and b play S1 and S2 simultaneously, respectively. In segments C and D, S1 is out of phase and S2 is in phase. a2: During each playback of the aforementioned stimulus signal, a miniature low-noise microphone is used to simultaneously collect the cochlear response in segments A, B, C, and D of the ear canal, which are recorded sequentially as: p A (t), p B (t), p C (t), p D (t); p A In addition to background noise, it also contains positive-phase stimulus artifacts. and the SFOAEs signal SFE induced by it + ; p B In addition to background noise, it also contains inverse stimulus sound artifacts. and the inverted SFOAEs signal SFE induced by it - ; p C It mainly contains stimulus sound artifacts Suppressing acoustic artifacts SFOAEs and SFEs induced by S2 + ; p D The main components include antiphase stimulus artifacts. Suppression of acoustic artifacts in positive phase SFOAEs and SFEs induced by S2 + ; a3: The corresponding signal in the ear canal collected by the miniature low-noise microphone is passed through a bandpass filter with a bandwidth of f1 to f2 Hz to eliminate high-frequency and low-frequency background noise interference; where f2 > f1. a4: The calculated time-domain swept-frequency SFOAEs residual signal 2p SFOAE (t), the specific calculation method is as follows: a5: Use a dynamic tracking filter to process the obtained residual signal 2p. SFOAE (t) Perform dynamic tracking filtering to remove the SFOAEs signal of the sweep frequency suppression sound S2 that has not been completely canceled, and obtain the target SFOAEs signal. S3: Process the target SFOAEs signal to obtain time-domain and frequency-domain information, and calculate the amplitude spectrum and phase spectrum of the swept-frequency SFOAEs signal based on the time-frequency distribution diagram; Specifically, the following steps are included: b1: Use IFFT to transform the target SFOAEs signal of the complex spectrum from the frequency domain to the time domain or latency domain, and obtain the time-frequency distribution map of SFOAEs based on CWT; b2: Based on least squares fitting, a filter cutoff line applicable to all stimulus intensities is obtained from the time-frequency distribution plot of SFOAEs; b3: Based on the different ear canal positions, different extraction targets are set. Based on the fitted filter f-τ cutoff line function, the wavelet coefficients of the latent components to be filtered out are set to zero on the time-frequency distribution plot to obtain the target wavelet coefficient components. b4: The target wavelet coefficient components are calculated using inverse CWT to obtain the time-domain signal of the SOFEs components; b5: The time-domain signal of the SOFEs components is processed by FFT operation to obtain the amplitude spectrum and phase spectrum of the SFOAEs components generated at different locations and with different latencies.

2. The source separation method for adaptive energy filtering based on swept-frequency SFOAEs according to claim 1, characterized in that: The scanning rate b is calculated as follows: Where, f1 = f(t1) f(t) = f1 + b(t - t1) f(t) is the instantaneous frequency of the stimulus signal, which increases linearly with time, and t1 and t2 are the start and end times of the scanning period.

3. The source separation method for adaptive energy filtering based on swept-frequency SFOAEs according to claim 1, characterized in that: In step a1, during each individual playback of the stimulus signal: Add 2 seconds of duration before the start of segment A. d The silent section, after section D, will have its duration increased by T. d The silent segment, in which T d The system delay is measured in advance.

4. The source separation method for adaptive energy filtering based on swept-frequency SFOAEs according to claim 1, characterized in that: In step a5, the dynamic tracking filter uses a real-time dynamic tracking filter whose center frequency changes continuously over time.

5. The source separation method for adaptive energy filtering based on swept-frequency SFOAEs according to claim 1, characterized in that: The fitted filter f-τ cutoff line function is: t i =c i ×a×f b Where f is the frequency of SFOAEs, τ is the latency of SFOAEs, i is the index of the filter cutoff line, and a, b, and c are the parameters to be determined for the fitting function.

Images