Active noise cancellation system and its control method, abnormal noise detection method and device

Abstract

This disclosure provides an active noise cancellation system and its control method, an abnormal noise detection method, and an apparatus. The abnormal noise detection method includes: acquiring a first noise signal during the operation of the active noise cancellation system; the first noise signal includes noise signals from a noise source and / or control signals from the loudspeaker; performing a Fourier transform on the first noise signal, segmenting the spectrum obtained from the Fourier transform based on octave bands, and determining the power of each signal segment of the spectrum; and determining the presence of a stable single-frequency abnormal noise in response to a signal segment whose power difference with an adjacent signal segment is greater than a difference threshold. This disclosure, combining spectrum analysis and octave bands, can extract power features from the first noise signal that can accurately identify abnormal noise, thereby improving the accuracy of abnormal noise detection in the active noise cancellation system.

Description

Technical Field

[0001] This disclosure relates to the field of active noise cancellation technology, and in particular to an active noise cancellation system and its control method, abnormal sound detection method, and device. Background Technology

[0002] In active noise cancellation systems, the noise reduction system utilizes the superposition and interference of a speaker control signal (inverting noise) generated based on the input signal (including the acquired noise signal and feedback signal) and the noise signal to reduce the noise signal and achieve noise reduction. However, during this process, a howling phenomenon occurs when the feedback signal is greater than the noise signal. Even if the feedback signal is less than the noise signal, when the ratio of the feedback signal to the noise signal exceeds a certain threshold, some stable single-frequency abnormal sounds will still be emitted. The amplitude of these stable single-frequency abnormal sounds is much smaller than the noise during howling, and the amplitude of the speaker control signal generated when stable single-frequency abnormal sounds exist is not significantly different from the control signal generated when the system is operating normally. Therefore, traditional howling monitoring schemes based on the amplitude of the speaker control signal in existing technologies cannot effectively identify this stable single-frequency abnormal sound phenomenon. Summary of the Invention

[0003] The technical problem to be solved by this disclosure is to overcome the above-mentioned defects in the prior art and to provide an active noise reduction system and its control method, abnormal sound detection method and device.

[0004] This disclosure solves the above-mentioned technical problems through the following technical solution:

[0005] Firstly, a method for detecting abnormal noise is provided, applied to an active noise cancellation system, the active noise cancellation system including a microphone and a speaker; the method for detecting abnormal noise includes:

[0006] Acquire a first noise signal during the operation of the active noise cancellation system; the first noise signal includes noise signals from noise sources collected by the microphone and / or control signals from the speaker;

[0007] Perform a Fourier transform on the first noise signal, segment the spectrum obtained by the Fourier transform based on octave bands, and determine the power of each signal segment of the spectrum.

[0008] In response to the presence of a signal segment whose power difference with an adjacent signal segment is greater than the difference threshold, a stable single-frequency abnormal sound is determined to exist.

[0009] Optionally, the difference threshold is determined in the following way:

[0010] Obtain abnormal samples and normal samples; the abnormal samples are noise signals containing abnormal sounds, and the normal samples are noise signals without abnormal sounds;

[0011] Construct an objective function; the objective function is constructed based on at least one of the following parameters: the number of missed abnormal samples, the number of false positives of normal samples, the proportion of missed abnormal samples, and the proportion of false positives of normal samples;

[0012] The objective function is optimized within a preset optimization range of the difference threshold to determine the final difference threshold.

[0013] Optionally, the optimization algorithm includes at least one of the following algorithms: genetic algorithm, simulated annealing algorithm, particle swarm optimization algorithm, Bayesian optimization algorithm, and hill climbing algorithm;

[0014] And / or, the objective function is the ratio of the number of anomalies to the total number of samples, wherein the number of anomalies is the sum of the number of missed anomaly samples and the number of false positives among normal samples.

[0015] Optionally, a first noise signal during the operation of the active noise cancellation system is acquired at a target sampling frequency; the target sampling frequency is determined based on the heterophonic frequency and / or the computing power of the analysis device for the first noise signal; the heterophonic frequency is determined based on noise samples containing heterophonic sounds.

[0016] And / or, perform a Fourier transform on the first noise signal with a target frequency resolution; wherein the target frequency resolution is obtained through optimization, and the optimization objective is to accurately identify abnormal sounds while minimizing the computational load of abnormal sound detection.

[0017] Optionally, the spectrum obtained by Fourier transform is segmented based on octave bands, including: performing A-weighting on the spectrum and segmenting the A-weighting result.

[0018] And / or, in response to the existence of a signal segment whose power difference with an adjacent signal segment is greater than a difference threshold, determining the existence of a stable single-frequency abnormal sound includes: in response to the occurrence of a signal segment whose power difference with an adjacent signal segment is greater than a difference threshold more than a number of occurrences, determining that the active noise cancellation system has a stable single-frequency abnormal sound.

[0019] Secondly, a control method for an active noise reduction system is provided, including:

[0020] The active noise cancellation system is subjected to abnormal noise detection according to any one of the first aspects;

[0021] In response to the presence of a stable single-frequency abnormal sound in the active noise cancellation system, the active noise cancellation system is restarted.

[0022] Thirdly, a noise detection device is provided for use in an active noise cancellation system, the active noise cancellation system including a microphone and a speaker; the noise detection device is used to implement the noise detection method described in any one of the first aspects; the noise detection device includes:

[0023] The acquisition module is used to acquire a first noise signal during the operation of the active noise cancellation system; the first noise signal is a noise signal from a noise source collected by the microphone or a control signal from the speaker;

[0024] The calculation module is used to perform Fourier transform on the first noise signal, segment the spectrum obtained by Fourier transform based on octave bands, and determine the power of each signal segment of the spectrum.

[0025] The detection module determines the presence of a stable single-frequency abnormal sound in response to the presence of a signal segment whose power difference with the adjacent signal segment is greater than the difference threshold.

[0026] Fourthly, an active noise cancellation system is provided, including a memory, a processor, and a computer program stored in the memory and for running on the processor, wherein the processor executes the computer program to implement the method of any one of the first or second aspects.

[0027] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of this disclosure.

[0028] The positive and progressive effects of this disclosure are as follows: by combining spectrum analysis and octave band, this disclosure can extract power characteristics from the first noise signal that can accurately identify abnormal sounds, thereby improving the accuracy of abnormal sound detection in the active noise cancellation system. Attached Figure Description

[0029] Figure 1 A flowchart of an abnormal sound detection method provided as an exemplary embodiment of this disclosure;

[0030] Figure 2 A schematic diagram of frequency domain signal curves for a stable single-frequency abnormal sound, howling, and normal noise provided as an exemplary embodiment of this disclosure;

[0031] Figure 3 A schematic diagram of the time-domain signal curves of a stable single-frequency abnormal sound, howling, and normal noise provided for an exemplary embodiment of this disclosure;

[0032] Figure 4 A flowchart of a difference threshold optimization method in an abnormal sound detection method provided as an exemplary embodiment of this disclosure;

[0033] Figure 5 A schematic diagram comparing the curve effects of A-weighting processing in an abnormal sound detection method provided as an exemplary embodiment of this disclosure;

[0034] Figure 6 A schematic diagram of a noise detection device provided for an exemplary embodiment of this disclosure;

[0035] Figure 7 This is a schematic diagram of an active noise reduction system provided as an exemplary embodiment of the present disclosure. Detailed Implementation

[0036] The present disclosure is further illustrated below by way of embodiments, but the present disclosure is not limited to the scope of the embodiments described herein.

[0037] The prefixes such as "first" and "second" used in this disclosure are merely for distinguishing different descriptive objects and do not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes used to distinguish descriptive objects in this disclosure does not constitute a limitation on the described objects. The description of the described objects is given in the claims or the context of the embodiments, and should not be construed as an unnecessary limitation. Furthermore, in the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0038] Figure 1 The flowchart illustrates an abnormal noise detection method provided as an exemplary embodiment of this disclosure. The abnormal noise detection method is applied to an active noise cancellation system, which includes devices such as a speaker, a controller, and a microphone.

[0039] The abnormal noise detected in this embodiment refers to a noise emitted by the speaker of the active noise cancellation system due to certain faults or interference during operation, which affects the user experience. Abnormal noise includes howling and stable single-frequency abnormal noise (or stable single-frequency sound).

[0040] The characteristics of howling during active noise cancellation are: see Figure 2 and Figure 3 The noise spectrum exhibits a prominent peak, with the amplitude at its highest point significantly higher than adjacent frequencies. This peak is generally much larger than the peak of a stable single-frequency abnormal sound, accounting for a large portion of the total noise power. From the moment it is generated, the power amplitude in the time domain gradually increases, with the ratio of maximum power to minimum power generally exceeding 5 times, and the amplitude typically approaching or even exceeding the maximum output power of the loudspeaker itself. Figure 2 The second to fifth columns of the table refer to the amplitude (power) at 94.28Hz, 108.75Hz, 0Hz and 6000Hz respectively, and the sixth column, RMS data, refers to the root mean square value of the amplitude from 0 to 6000Hz.

[0041] Unlike howling, stable single-frequency abnormal noise / stable single-frequency sound is characterized by a prominent peak in the noise spectrum. The amplitude of the peak (noise peak value) is higher than the amplitude of the corresponding adjacent frequency, and the power of this peak accounts for a relatively small proportion of the total noise power. That is, the power of the peak accounts for a proportion of the total noise power that is less than or equal to the power threshold, where the power threshold is determined based on experimental data. Figure 2 As shown by the blue curve, the bandwidth of this peak does not exceed 50Hz. See also Figure 3 In stable single-frequency noise, the power amplitude in the time domain is relatively stable from the beginning of its generation, with little power variation. The ratio of the maximum power to the minimum power will not exceed 3 times, and the amplitude is generally much smaller than the maximum sound power of the speaker itself. Figure 3 The left vertical axis represents the power amplitude, and the right vertical axis represents the sound pressure amplitude; the two can be converted to each other.

[0042] This disclosure primarily detects whether a stable single-frequency abnormal sound occurs in the active noise cancellation system. (See also...) Figure 1 The abnormal sound detection method includes the following steps:

[0043] Step 101: Obtain the first noise signal during the operation of the active noise cancellation system.

[0044] The first noise signal can be the noise signal emitted by the noise source collected by the first microphone. This noise signal will be used as the basis for abnormal sound detection. The first microphone is deployed near the noise source. It should be noted that the first noise signal can be the result of a single first microphone or a combination of results from multiple first microphones. In this case, the first noise signal collected by the first microphone is the superposition of the original noise signal emitted by the noise source and the feedback signal.

[0045] Active noise cancellation systems can be used to reduce noise in appliances such as range hoods and washing machines. Taking the noise reduction of a range hood as an example, the noise source can be, for example, the range hood's fan, and a first microphone for collecting noise signals is deployed near the fan.

[0046] The first noise signal can be the speaker's control signal, which will be used as the basis for subsequent abnormal noise detection. It should be noted that the speaker's control signal can be obtained directly from the controller's output, or it can be acquired through a second microphone deployed near the speaker. The first noise signal can be the control signal of a single speaker, or a combination of control signals from multiple speakers.

[0047] The first noise signal can also be a combination of the noise signal from the noise source and the control signal from the loudspeaker. The specific combination method is not particularly limited in the embodiments disclosed herein.

[0048] In one embodiment, the first noise signal meets the sampling frequency requirements and total sampling time requirements for Fourier transform processing, such as the sampling frequency f. s ≥2f U , where f U The highest analysis frequency (e.g., 12800Hz) of the active noise reduction system during noise signal acquisition, where the highest analysis frequency is greater than or equal to the highest possible frequency f of the abnormal noise. ANC For example, abnormal sounds may occur at frequencies of 700Hz, 800Hz, and 900Hz. ANC 900Hz, f U ≥900Hz. Total sampling time T ≥ duration of a single Fourier transform T FFT T FFT The frequency resolution Δf required for abnormal sound detection is determined, where Δf ≤ 100Hz. Preferably, Δf ≤ 20Hz.

[0049] The highest analysis frequency for active noise reduction refers to the target control frequency. Active noise reduction only analyzes noise within a specific frequency range and does not control noise outside that range. The maximum analysis frequency for active noise reduction control ranges from [600Hz to 2000Hz]. Preferably, the maximum analysis frequency for active noise reduction control is [600Hz to 1000Hz].

[0050] For example, in a scenario where an active noise cancellation system is used to reduce the noise of a range hood, the noise of the range hood is distributed in the range of 0 to 20000 Hz, while the frequency range of the analysis frequency of active noise cancellation may only be 0 to 1000 Hz. Correspondingly, the abnormal noise caused by the active noise cancellation system can only be in the range of 0 to 1000 Hz. Therefore, setting the highest analysis frequency of active noise cancellation to 1000 Hz is sufficient.

[0051] In one embodiment, a first noise signal during the operation of the active noise cancellation system is acquired at a target sampling frequency. The target sampling frequency is determined based on the abnormal sound frequency and / or the computing power of the analysis device for the first noise signal; the abnormal sound frequency is determined based on noise samples containing abnormal sounds. The noise samples can be obtained experimentally or by superimposing abnormal sounds onto noise samples without abnormal sounds. The analysis device for the first noise signal is the device that analyzes the first noise signal to achieve abnormal sound detection, and may be, for example, an active noise cancellation chip.

[0052] The following describes one method for determining the target sampling frequency:

[0053] The method for determining the minimum target sampling frequency: By analyzing noise samples containing abnormal sounds, it can be seen that the maximum frequency (abnormal sound frequency) of the abnormal sound is generally f. ANCAccording to the Nyquist sampling theorem, to avoid aliasing, the target sampling frequency should be at least twice the frequency of the different audio frequencies. Therefore, the minimum value of the target sampling frequency f is... s min =2f ANC .

[0054] The method for determining the maximum value of the sampling frequency: Analyzing the first noise signal includes performing a Fourier transform on it. The computing power of the analysis device can be characterized by the number of data points of the first noise signal in each Fourier transform. The maximum value of the target sampling frequency can be derived from the maximum computing power. Assuming the analysis device allows N data points to be transformed in one Fourier transform, the time of one Fourier transform is 1 / Δf, and the sampling frequency f... s Indicates f per second s Therefore, the number of data points allowed for a single Fourier transform is f. s ·1 / Δf, i.e., N=f s ·1 / Δf, assuming the maximum computing power of the analysis device is N and the maximum number of data points for each Fourier transform. max Then we can get f s_max =N max ·Δf.

[0055] Therefore, the target sampling frequency f of the first noise signal s_res The range of values ​​for is [f s_min f s_max In related technologies, to improve the control accuracy of active noise cancellation, the original signal acquired by the active noise cancellation controller has a high sampling frequency f. s , usually f s The target sampling frequency is 12000Hz. In this embodiment, the target sampling frequency f is less than 12000Hz. s_res Obtain the first noise signal.

[0056] In this embodiment, the first noise signal is obtained at the target sampling frequency determined by the computing power of the analysis device based on the abnormal sound frequency and / or the first noise signal. This serves as the data basis for abnormal sound detection. On the one hand, it can provide a data basis that conforms to the characteristics of abnormal sound for subsequent steps, eliminate interference, and thus improve the accuracy of abnormal sound detection. On the other hand, the first noise signal obtained at the target sampling frequency eliminates some interference signals, reduces the amount of data analysis, and thus improves the abnormal sound detection rate and meets the timeliness requirements.

[0057] Step 102: Perform Fourier transform on the first noise signal, segment the spectrum obtained by Fourier transform based on octave bands, and determine the power of each signal segment of the spectrum.

[0058] In addition to the 1 / n octave band method, segmentation algorithms can also employ, but are not limited to, the Bark bandwidth method and / or the equal bandwidth method. Here, n in the 1 / n octave band method is a non-zero natural number such as 1, 2, or 3. To balance detection accuracy and computational power, n=3 is preferred.

[0059] Performing a Fourier transform on the first noise signal yields its spectrum (Fourier transform result). Spectral analysis reveals that when howling and stable single-frequency abnormal sounds occur, a signal peak significantly higher than the surrounding frequencies exists in the spectrum. Figure 2 The peak frequency near 95Hz (dotted red line) and the peak frequency near 109Hz (thin solid blue line) are shown. Through filtered playback, it was found that the noise corresponding to these peak frequencies is the sound that produces the howling and stable single-frequency abnormal noise. The normal signal (…) Figure 2 (The thick green solid line in the image) does not exhibit this obvious, abrupt noise peak. Furthermore, the octave-based segmentation strategy allows for a precise analysis of the frequency distribution of noise signal power. Therefore, the combination of spectral analysis and octave segmentation can extract power characteristics that accurately characterize stable single-frequency abnormal sounds.

[0060] In one embodiment, to improve the algorithm speed, a Fast Fourier Transform (FFT) is used to perform a Fourier transform on the first noise signal.

[0061] In one embodiment, the number of signal data points selected for a single FFT process is related to T. FFT / f s_res The value is close to 2 N N is a positive integer, and it is guaranteed that the selected N satisfies 2. N ≥T FFT / f s_res The final spectrum data D FFT Choose either the result of a single FFT or the power average of multiple FFT results.

[0062] In one embodiment, a Fourier transform is performed on the first noise signal at a target frequency resolution, and the active noise reduction system is used to detect abnormal sounds based on the Fourier transform result; wherein, the target frequency resolution is obtained through optimization, and the optimization objective is to accurately identify abnormal sounds while minimizing the computational load of abnormal sound detection.

[0063] Accurate identification of abnormal sounds refers to an accuracy rate greater than the accuracy threshold and / or a false identification rate less than or equal to the false identification threshold. The accuracy threshold and false identification threshold can be set according to the specific circumstances.

[0064] Frequency resolution determines the computational load of the abnormal sound detection process. When the frequency resolution is 1Hz, it means that a Fourier transform is performed every 1 second. Assuming that the computational load for each operation is K1, the computational load for each operation is K2, which is the amount of data within 1 / 4 second. K1 = 4·K2. The amount of data with a frequency resolution of 4Hz is reduced by 4 times compared to a frequency resolution of 1Hz, which greatly speeds up the computational speed of abnormal sound detection.

[0065] As the frequency resolution increases, the amplitude of the Fourier transform amplitude-frequency graph increases. In other words, when the frequency resolution increases, the number of spectral lines decreases, and the power of each spectral line is the sum of the powers of the spectral lines before the frequency resolution was increased.

[0066] The following example further illustrates the process of determining the target frequency resolution:

[0067] Based on the abnormal sound samples, the frequency bands in which the abnormal sounds occur are known, such as the frequencies fa, fb, and fc in the abnormal sound samples, which contain stable single-frequency abnormal sounds.

[0068] S1. Assuming the initial frequency resolution is 1Hz, after increasing the frequency resolution to 32Hz, perform FFT on the abnormal sound samples and detect abnormal sounds based on the FFT.

[0069] S2. If a frequency resolution of 32Hz can accurately identify stable single-frequency abnormal sounds in the fa, fb, and fc frequencies of the abnormal sound sample, then increasing the frequency resolution to 32Hz is sufficient to detect the abnormal sounds. If a frequency resolution of 32Hz cannot accurately identify abnormal sounds in all frequencies fa, fb, and fc of the abnormal sound sample and / or there are misjudgments (e.g., identifying other frequencies as abnormal sounds), then although increasing the frequency resolution to 32Hz reduces the computational load, it is insufficient to detect all abnormal sound points, and the frequency resolution needs to be reduced.

[0070] S3. If the frequency resolution is sufficient to identify abnormal sounds, select 32Hz as the frequency resolution and determine the corresponding computational load. At this point, the computational load has been reduced to about 1 / 32 compared to the initial computational load when the frequency resolution is 1Hz. If the frequency resolution is insufficient to identify all active noise reduction abnormal sound points, continue to repeat step 2 until the frequency resolution that accurately identifies abnormal sounds and has the smallest computational load for abnormal sound detection is found.

[0071] In this embodiment, the first noise signal is subjected to spectrum analysis based on the target frequency resolution obtained by optimization, which can provide a data basis that conforms to the characteristics of abnormal sounds, eliminate interference, and thus improve the accuracy of abnormal sound detection; at the same time, the amount of computation for abnormal sound detection is minimized, thereby improving the abnormal sound detection rate and meeting the timeliness requirements.

[0072] Step 103: In response to the existence of a signal segment whose power difference with the adjacent signal segment is greater than the difference threshold, it is determined that there is a stable single-frequency abnormal sound.

[0073] In step 103, the adjacent signal segments of signal segment i are signal segment i-1 and signal segment i+1. If the power P of signal segment i... i The power P of signal segment i-1 i-1 The difference is greater than or equal to the difference threshold, and the power P of signal segment i is greater than or equal to the difference threshold. i The power P of signal segment i+1 i+1 If the difference is greater than or equal to the difference threshold, then a stable single-frequency abnormal sound is determined to exist. Otherwise, it is determined that the active noise cancellation system does not have a stable single-frequency abnormal sound.

[0074] In this embodiment, by performing spectrum analysis and octave band processing on the first noise signal, the power characteristics that can accurately characterize stable single-frequency abnormal sounds can be extracted. Based on the power characteristics, abnormal sounds can be accurately identified, thereby improving the abnormal sound detection accuracy of the active noise cancellation system.

[0075] The difference threshold in step 103 can be an empirical value, or it can be dynamically selected based on actual conditions to improve the accuracy of abnormal sound detection. Below is a method for dynamically selecting the difference threshold; see [link to relevant documentation]. Figure 4 The difference threshold is determined in the following way:

[0076] Steps 100-11: Obtain abnormal signal samples and normal samples.

[0077] Abnormal samples are noise signals containing abnormal sounds, while normal samples are noise signals without abnormal sounds. Abnormal samples can be collected experimentally or are the result of superimposing normal samples with noise signals whose frequency and amplitude match those of the abnormal sounds.

[0078] The number of abnormal and normal samples can be set according to actual needs. Understandably, the more samples there are, the better the difference threshold can be optimized.

[0079] Steps 100-12: Construct the objective function.

[0080] The objective function is constructed based on at least one of the following parameters: the number of missed abnormal samples, the number of false positives among normal samples, the percentage of missed abnormal samples, and the percentage of false positives among normal samples.

[0081] The design principle of the objective function is to find a combination and value of judgment parameters that allows the optimized difference threshold to detect all abnormal sounds while preventing normal signals from being misjudged as abnormal sounds, and also to have good robustness. The objective function for parameter optimization can be set as: (1) (number of missed abnormal samples + number of false normal samples) / total number of samples, where the smaller the objective function, the better; (2) (abnormal sound judgment intensity of normal samples) / (abnormal sound judgment intensity of abnormal signals), where the smaller the objective function, the better.

[0082] The abnormal sound determination intensity is calculated as: Abnormal Sample / Normal Sample. If the total abnormal sound threshold is 100 (anything above 100 is considered abnormal), then if the normal sample value is 30, the abnormal sound intensity is 0.3. This value represents the abnormal sound determination intensity of the normal sample. The smaller this value, the further the normal sound deviates from the abnormal sound standard, the higher the probability of it being a normal sound, and the higher the accuracy of the detection algorithm. If the abnormal sound sample value is 200, the abnormal sound intensity is 2. This value represents the abnormal sound determination intensity of the abnormal signal. The larger this value, the more likely the sound is to be abnormal, the higher the probability of it being abnormal, and the higher the accuracy of the detection algorithm. Therefore, the smaller the value of (abnormal sound determination intensity of normal samples) / (abnormal sound determination intensity of abnormal signals), the higher the accuracy of the detection algorithm, and the better the corresponding parameter combination used for detection.

[0083] The intensity of abnormal sound determination can also be defined in other ways, for example: the intensity of abnormal sound determination for normal samples = the number of normal samples that are misidentified as abnormal sounds / the total number of normal samples; the intensity of abnormal sound determination for abnormal signals = the number of abnormal samples that are accurately determined as abnormal sounds / the total number of abnormal samples.

[0084] In addition to the objective function mentioned above, other similar calculation formulas that conform to the design principles of objective functions can also be used, and this disclosure does not impose any particular limitations.

[0085] Steps 100-13: Optimize the objective function within the preset optimization range of the difference threshold to determine the final difference threshold.

[0086] In one embodiment, the preset optimization range is [1dB, 10dB], that is, the difference threshold is used to find the optimal solution in [1dB, 10dB] so that the objective function is minimized.

[0087] In this embodiment, an optimization range for the difference threshold is provided, so that the optimization conforms to the physical characteristics of certain abnormal sounds, thereby improving the optimization accuracy and efficiency.

[0088] In one embodiment, the optimization algorithm includes at least one of the following algorithms: genetic algorithm, simulated annealing algorithm, particle swarm optimization algorithm, Bayesian optimization algorithm, and hill climbing algorithm.

[0089] In one embodiment, the step of segmenting the spectrum obtained by Fourier transform based on octave bands includes: performing A-weighting on the spectrum and segmenting it according to the A-weighting result.

[0090] Currently, the definition of abnormal sounds is mainly based on human subjective perception. Therefore, although some frequencies may be loud, they can still be ignored if the human ear is not sensitive to sounds in this frequency band (such as low-frequency sounds). In the abnormal sound detection process, the Fourier transform result of the first noise signal is first processed using A-weighting based on the sensitivity of the human ear. Then, the abnormal sound detection is performed on the A-weighted result. Since A-weighting can better reflect human auditory perception, this detection method is more likely to detect real abnormal sounds, resulting in better detection results and a better match with the user's perception and judgment of abnormal sounds.

[0091] See Figure 5 The solid green line represents the spectrum of the first noise signal before A-weighting, which contains a large number of low-frequency components below 100Hz that are insensitive to the human ear. The sound signal after A-weighting of the first noise signal is as follows: Figure 5 As shown by the red dotted line, the proportion of total power in low-frequency components, which are less sensitive to human hearing, decreases. Using the A-weighted spectrum for abnormal sound detection results in more accurate detection and better reflects human auditory perception.

[0092] It should be noted that since it is easier to identify abnormal sounds after A-weighting, the difference threshold after A-weighting can be smaller than the difference threshold without A-weighting. For example, the difference threshold without A-weighting is 5dB, while the difference threshold after A-weighting is 4dB.

[0093] In one embodiment, abnormal noise is detected by multiple judgments. Step 103 includes: in response to the occurrence of a signal segment whose power difference with an adjacent signal segment is greater than a difference threshold, the number of occurrences is greater than a number threshold, and it is determined that there is a stable single-frequency abnormal noise in the active noise cancellation system.

[0094] When detecting abnormal noise using the spectral data of the first noise signal, if the acquired first noise signal contains noise interference signals, or if the first noise signal used for detection is not a completely steady-state signal but exhibits certain fluctuation characteristics over time, these factors can lead to differences in the spectral data obtained at different times, resulting in misjudgments or incorrect detection of abnormal noise.

[0095] To reduce or avoid misjudgments, in this embodiment, whenever the difference between the total amplitude of the spectral power and the peak noise value is less than the difference threshold, it is determined that the active noise cancellation system is abnormal, and the number of abnormal occurrences is incremented by 1. When the number of occurrences is greater than the number threshold, it is determined that the active noise cancellation system has a stable single-frequency abnormal sound. By making multiple judgments, misjudgments can be effectively reduced or avoided.

[0096] The number of detections threshold can be set according to actual needs. It can be an empirical value, such as 3 times; or it can be dynamically selected through target optimization to improve the accuracy of abnormal sound detection. The target optimization process for the number of detections threshold is similar to that for the difference threshold, and will not be elaborated here.

[0097] Many abnormal sound frequencies fluctuate, and these fluctuations significantly affect the amplitude of the corresponding frequency. Therefore, using the frequency spectrum for abnormal sound identification can lead to false positives and false negatives. Furthermore, human perception and judgment of abnormal sounds are based on the overall power within a frequency band, rather than the power at a single frequency. Therefore, integrating the noise spectrum according to a certain rule and using the power of the corresponding frequency band after integration to determine abnormal sounds is more consistent with human perception.

[0098] Based on this, in one embodiment, the Fourier transform result (spectrum) of the first noise signal is segmented to obtain at least two noise segments, and the power is determined based on the integral of each noise segment.

[0099] In addition to the 1 / n octave band method, segmentation algorithms can also employ, but are not limited to, the Bark bandwidth method and / or the equal bandwidth method. Here, n in the 1 / n octave band method is a non-zero natural number such as 1, 2, or 3. To balance detection accuracy and computational power, n=3 is preferred.

[0100] Devices equipped with active noise cancellation systems, such as range hoods, may also produce sounds with spectral characteristics similar to the stable single-frequency tone caused by the active noise cancellation system. We call this "normal abnormal noise." Although this "normal abnormal noise" can also be annoying to users, it is different from the stable single-frequency tone caused by the active noise cancellation system. It is not a "byproduct" of the active noise cancellation system and cannot be reduced by turning off or adjusting the active noise cancellation system.

[0101] Based on this, in one embodiment, if the detection result of step 103 is that there is a stable single-frequency abnormal sound, in order to rule out that the abnormal sound is caused by the device equipped with the active noise cancellation system rather than by the active noise cancellation system, the active noise cancellation system is first turned off, and a second noise signal is obtained during the process of turning off the active noise cancellation system. Steps 101 to 103 are then performed on the second noise signal. If the determination result is that there is a stable single-frequency abnormal sound in the second noise signal, and the abnormal sound frequency of the second noise signal is the same as or similar to that of the first noise signal, then it is determined that the abnormal sound is caused by the device rather than by the active noise cancellation system; otherwise, it is determined that there is a stable single-frequency abnormal sound in the active noise cancellation system.

[0102] The following provides another method to eliminate the interference of device noise on the noise detection of the active noise cancellation system.

[0103] In one embodiment, see Figure 6 Before acquiring the first noise signal during the operation of the active noise cancellation system, the process also includes:

[0104] Step 100: Without the active noise cancellation system being activated, acquire the second noise signal of the noise source and segment the second noise signal so that the power proportion of each noise segment of the second noise signal is less than the proportion threshold, and determine the segmentation strategy of the second noise signal.

[0105] The second noise signal is collected through the first microphone.

[0106] The number of noise segments obtained by segmenting the first noise signal is determined based on the power proportion of each noise segment. If the power proportion of each noise segment is not less than the proportion threshold, further segmentation is required. The frequency bands of the noise segments can be the same or different. The second noise signal is segmented so that the power proportion of each noise segment is less than the proportion threshold, that is, the high-power frequency band of the range hood itself is separated out, thereby eliminating the influence of the range hood's own frequency band characteristics on the abnormal noise judgment.

[0107] The percentage threshold can be set according to the actual situation, and the range of the percentage threshold can be, for example, [40%, 50%]. Preferably, the percentage threshold is 40%. Choosing 40% is to leave a margin and avoid misjudgment that might occur if it is too close to 50%. This is because: since the subsequent criterion for judging abnormal noise is that exceeding the percentage threshold of 50% is considered to indicate the presence of a stable single-frequency abnormal noise, and since the power percentage of a certain noise segment cannot be guaranteed to be an accurate value, there may be fluctuations of up to 5%. Therefore, if it is too close to 50%, it may cause misjudgment of abnormal noise due to fluctuation errors. Setting the percentage threshold to 40%, when the power percentage is less than 40%, indicates that the power of the noise segment has been reduced low enough to avoid misjudgment caused by power fluctuations in the noise segment.

[0108] The segmentation strategy includes the following parameters: number of segments, upper limit frequency and lower limit frequency of each noise segment.

[0109] The following describes how to calculate the power percentage, using the following formula:

[0110] H = 1 - e i / E;

[0111]

[0112] Where H represents the power percentage; e i X represents the power corresponding to the i-th noise segment in the noise signal; E represents the sum of the power of all noise segments contained in the noise signal; j (t) represents the amplitude of the j-th sampling point of the noise signal in the time domain; N represents the number of noise segments at the sampling points contained in the noise signal. i The calculation method is similar to that of E.

[0113] In step 102, the first noise signal is segmented according to a segmentation strategy, and a Fourier transform is performed on the resulting noise segments, or the Fourier transform results are segmented according to the segmentation strategy. Subsequently, the global total spectral power amplitude and noise peak value of the first noise signal are determined based on the integral of each noise segment, or the total spectral power amplitude and noise peak value of each noise segment are determined separately, i.e., the local total spectral power amplitude and noise peak value.

[0114] In this embodiment, abnormal noise from the active noise reduction system can be quickly detected, eliminating the influence of the range hood's own frequency band characteristics on the abnormal noise judgment, and achieving accurate monitoring of abnormal noise from the range hood's active noise reduction system.

[0115] The following provides another method to eliminate the interference of device noise on the noise detection of the active noise cancellation system.

[0116] In one embodiment, before acquiring the first noise signal during the operation of the active noise cancellation system, the method further includes:

[0117] Steps 100-21: Acquire the second noise signal of the noise source when the active noise cancellation system is not activated.

[0118] Steps 100-22: Perform octave band calculations on the second noise signal to determine a segmentation strategy that matches the second noise signal.

[0119] In 1 / n octave bands, n is a non-zero natural number such as 1, 2, or 3. To balance detection accuracy and computing power, n=3 is preferred.

[0120] Steps 100-23: Perform segmentation processing on the second noise signal according to the segmentation strategy, and calculate the characteristic value of each noise segment of the second noise signal.

[0121] Among them, the characteristic values ​​include power and / or decibel values.

[0122] Steps 100-24: Determine the target gain of the first target noise segment whose eigenvalue is greater than the eigenvalue threshold.

[0123] After filtering the first target noise segment with a bandpass filter of the target gain, the eigenvalue of the first target noise segment is less than or equal to the eigenvalue threshold.

[0124] The eigenvalue of the first target noise segment is greater than the eigenvalue threshold, indicating that the first target noise segment contains abnormal noise. Since the active noise cancellation system was not activated when the first target noise segment was collected, this means that the abnormal noise contained in the first target noise segment is caused by the device equipped with the active noise cancellation system, rather than by the active noise cancellation system itself. To eliminate device-induced abnormal noise interference during abnormal noise detection, it is necessary to filter the second target noise segment contained in the first noise signal to filter out the device-induced abnormal noise. The frequency band of the second target noise segment includes the frequency band of the first target noise segment.

[0125] Taking the characteristic value as a decibel value as an example, assuming the decibel value of the first target noise segment before adjustment is P, and the decibel value after adjustment is P' (P' is less than or equal to the characteristic value threshold), here P' is set to be 1 dB greater than the adjacent frequency band decibel value. According to the formula: P'=20lg(E' / E0), the power of the first target noise segment after adjustment is E'. Using the known power E before adjustment and power E' after adjustment of the first target noise segment, the target gain / pass rate of the first target noise segment is calculated as α=E' / E.

[0126] It should be noted that P' is set to be 1 dB greater than the adjacent frequency band value, rather than equal to it, because this makes the abnormal noise detection more sensitive. Compared with the former, the former is more sensitive to abnormal noise. When the judgment condition is 3 dB higher than the adjacent frequency band, because it is 1 dB higher than the adjacent frequency band, less power is needed to judge the abnormal noise compared with the same value as the adjacent frequency band. At this time, the abnormal noise judgment sensitivity is higher while ensuring accuracy.

[0127] In step 102, the first noise signal is segmented to obtain at least two noise segments, and a bandpass filter with target gain is used to filter the second target noise segment contained in the first noise signal. Fourier transforms are then performed on the segmented noise segments and the filtered noise segments in the first noise signal, respectively.

[0128] Among them, the noise segments obtained by segmenting the first noise signal are the noise segments other than the second target noise segment.

[0129] A bandpass filter with the target gain is used to filter the second target noise segment, that is, α is multiplied by the amplitude of the second target noise segment to filter out the abnormal noise caused by the equipment in the second target noise segment. The amplitude of other noise segments in the second noise signal other than the second target noise segment is not attenuated, that is, the power pass rate of other noise segments is 100%.

[0130] In this embodiment, a bandpass filter with target gain is used to filter the noise segment of the target noise segment in the first noise signal, which can eliminate the interference of equipment noise on the noise detection of the active noise reduction system and further improve the accuracy of noise detection.

[0131] This disclosure also provides a control method for an active noise cancellation system, the method comprising: detecting abnormal noise in the active noise cancellation system according to the abnormal noise detection method provided in any of the above embodiments; and restarting the active noise cancellation system in response to the presence of a stable single-frequency abnormal noise in the active noise cancellation system.

[0132] Experience suggests that unusual noises in active noise cancellation systems are caused by certain malfunctions or interference. Restarting the active noise cancellation system can weaken or eliminate the malfunctions or interference, improve the effectiveness of the active noise cancellation system, and thus improve the user experience.

[0133] Corresponding to the aforementioned embodiments of the abnormal noise detection method, this disclosure also provides embodiments of the abnormal noise detection device.

[0134] Figure 6 A schematic diagram of an abnormal noise detection device provided for an exemplary embodiment of this disclosure, the device comprising:

[0135] The acquisition module 61 is used to acquire a first noise signal during the operation of the active noise cancellation system; the first noise signal includes the noise signal of the noise source and / or the control signal of the speaker;

[0136] The calculation module 62 is used to perform Fourier transform on the first noise signal, segment the spectrum obtained by Fourier transform based on octave bands, and determine the power of each signal segment of the spectrum.

[0137] The detection module 63 is used to determine the presence of a stable single-frequency abnormal sound in response to the presence of a signal segment whose power difference with the adjacent signal segment is greater than the difference threshold.

[0138] Optionally, the device further includes the difference threshold determination module, used for:

[0139] Obtain abnormal samples and normal samples; the abnormal samples are noise signals containing abnormal sounds, and the normal samples are noise signals without abnormal sounds;

[0140] Construct an objective function; the objective function is constructed based on at least one of the following parameters: the number of missed abnormal samples, the number of false positives of normal samples, the proportion of missed abnormal samples, and the proportion of false positives of normal samples;

[0141] The objective function is optimized within a preset optimization range of the difference threshold to determine the final difference threshold.

[0142] Optionally, the preset optimization range is [1dB, 10dB];

[0143] And / or, the optimization algorithm includes at least one of the following algorithms: genetic algorithm, simulated annealing algorithm, particle swarm optimization algorithm, Bayesian optimization algorithm, hill climbing algorithm;

[0144] And / or, the objective function is the ratio of the number of anomalies to the total number of samples, wherein the number of anomalies is the sum of the number of missed anomaly samples and the number of false positives among normal samples.

[0145] Optionally, if the spectrum of the first noise signal contains at least two peaks, the largest peak is determined as the noise peak.

[0146] And / or, acquire a first noise signal during the operation of the active noise cancellation system at a target sampling frequency; the target sampling frequency is determined based on the heterophonic frequency and / or the computing power of the analysis device for the first noise signal; the heterophonic frequency is determined based on noise samples containing heterophonic sounds;

[0147] And / or, perform a Fourier transform on the first noise signal with a target frequency resolution; wherein the target frequency resolution is obtained through optimization, and the optimization objective is to accurately identify abnormal sounds while minimizing the computational load of abnormal sound detection.

[0148] Optionally, the determining module is specifically used to: perform A-weighting processing on the spectrum and segment the A-weighting processing result;

[0149] And / or, the determination module is specifically used to: divide the Fourier transform result into at least two noise segments, and determine the total amplitude of the spectral power and the noise peak value based on the integral of each noise segment;

[0150] And / or, the detection module is specifically used to: determine that the active noise cancellation system has a stable single-frequency abnormal sound in response to the occurrence of a signal segment whose power difference with an adjacent signal segment is greater than a difference threshold being greater than a number threshold.

[0151] Optionally, the acquisition module is further configured to acquire a second noise signal from the noise source when the active noise cancellation system is not activated;

[0152] The device further includes: a segmentation module, used to segment the second noise signal so that the power proportion of each noise segment of the second noise signal is less than the proportion threshold, and to determine the segmentation strategy of the second noise signal;

[0153] The determination module is specifically used to: segment the first noise signal according to the segmentation strategy, and perform Fourier transform on the segmented noise segments.

[0154] Optionally, the acquisition module is further configured to acquire a second noise signal from the noise source when the active noise cancellation system is not activated;

[0155] The device also includes: a calculation module for performing octave band calculation on the second noise signal to determine a segmentation strategy that matches the second noise signal;

[0156] The segmentation module is used to segment the second noise signal according to the segmentation strategy and calculate the characteristic values ​​of each noise segment of the second noise signal; the characteristic values ​​include power and / or decibel values;

[0157] A gain module is used to determine the target gain of a first target noise segment whose eigenvalue is greater than a eigenvalue threshold; after filtering the first target noise segment with a bandpass filter of the target gain, the eigenvalue of the first target noise segment is less than or equal to the eigenvalue threshold;

[0158] The module is specifically used for:

[0159] The first noise signal is segmented to obtain at least two noise segments, and a bandpass filter with a target gain is used to filter the second target noise segment contained in the first noise signal. Fourier transforms are performed on the segmented noise segments and the filtered noise segments in the first noise signal, respectively; wherein the frequency band of the second target noise segment includes the frequency band of the first target noise segment.

[0160] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs.

[0161] Figure 7 This is a schematic diagram of an active noise cancellation system according to an example embodiment of the present disclosure. The active noise cancellation system includes a memory, a processor, and a computer program stored in the memory and used to run on the processor. When the processor executes the computer program, it implements the abnormal sound detection method or control method described in any of the above embodiments. Figure 7 The active noise cancellation system 70 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0162] like Figure 7 As shown, the components of the active noise cancellation system 70 may include, but are not limited to: at least one processor 71, at least one memory 72, and a bus 73 connecting different system components (including memory 72 and processor 71).

[0163] Bus 73 includes a data bus, an address bus, and a control bus.

[0164] The memory 72 may include volatile memory, such as random access memory (RAM) 721 and / or cache memory 722, and may further include read-only memory (ROM) 723.

[0165] The memory 72 may also include a program tool 725 (or utility) having a set (at least one) program module 724, such program module 724 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0166] The processor 71 executes various functional applications and data processing by running computer programs stored in the memory 72, such as the abnormal sound detection method or control method provided in any of the above embodiments.

[0167] The active noise cancellation system 70 can also communicate with one or more external devices 74 (e.g., range hood processors). This communication can be made via input / output (I / O) interface 75. Furthermore, the active noise cancellation system 70 can also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via network adapter 76. As shown, network adapter 76 communicates with other modules of the active noise cancellation system 70 via bus 73. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with the active noise cancellation system 70, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0168] It should be noted that although several units / modules or sub-units / modules of the active noise cancellation system have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0169] This disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the abnormal sound detection method or control method provided in any of the above embodiments.

[0170] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0171] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described abnormal sound detection or control methods.

[0172] The program code for executing the computer program product of this disclosure can be written in any combination of one or more programming languages, and the program code can be executed entirely on a user device, partially on a user device, as a stand-alone software package, partially on a user device and partially on a remote device, or entirely on a remote device.

[0173] While specific embodiments of this disclosure have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of this disclosure is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of this disclosure, but all such changes and modifications fall within the scope of protection of this disclosure.

Claims

1. A method for detecting abnormal sounds, characterized in that, The method is applied to an active noise cancellation system, which includes a microphone and a speaker; the abnormal sound detection method includes: Acquire a first noise signal during the operation of the active noise cancellation system; the first noise signal includes noise signals from noise sources collected by the microphone and / or control signals from the speaker; Perform a Fourier transform on the first noise signal, segment the spectrum obtained by the Fourier transform based on octave bands, and determine the power of each signal segment of the spectrum. In response to the presence of a signal segment whose power difference with an adjacent signal segment is greater than the difference threshold, a stable single-frequency abnormal sound is determined to exist. The characteristics of a stable single-frequency abnormal sound are: there is a prominent peak in the noise spectrum, the amplitude of the highest point of the peak is higher than the amplitude of the corresponding frequency of the adjacent frequency, and the power of the peak accounts for less than or equal to the power threshold of the total noise power.

2. The abnormal sound detection method according to claim 1, characterized in that, The difference threshold is determined in the following way: Obtain abnormal samples and normal samples; the abnormal samples are noise signals containing abnormal sounds, and the normal samples are noise signals without abnormal sounds; Construct an objective function; the objective function is constructed based on at least one of the following parameters: the number of missed abnormal samples, the number of false positives of normal samples, the proportion of missed abnormal samples, and the proportion of false positives of normal samples; The objective function is optimized within a preset optimization range of the difference threshold to determine the final difference threshold.

3. The abnormal sound detection method according to claim 2, characterized in that, The optimization algorithm includes at least one of the following algorithms: genetic algorithm, simulated annealing algorithm, particle swarm optimization algorithm, Bayesian optimization algorithm, and hill climbing algorithm; And / or, the objective function is the ratio of the number of anomalies to the total number of samples, wherein the number of anomalies is the sum of the number of missed anomaly samples and the number of false positives among normal samples.

4. The abnormal sound detection method according to claim 1, characterized in that, A first noise signal during the operation of the active noise cancellation system is acquired at a target sampling frequency; the target sampling frequency is determined based on the heterophonic frequency and / or the computing power of the analysis device for the first noise signal; the heterophonic frequency is determined based on noise samples containing heterophonic sounds.

5. The abnormal sound detection method according to claim 1, characterized in that, The first noise signal is subjected to a Fourier transform at a target frequency resolution; wherein the target frequency resolution is obtained through optimization, and the optimization objective is to accurately identify abnormal sounds while minimizing the computational load of abnormal sound detection.

6. The abnormal sound detection method according to claim 1, characterized in that, The spectrum obtained by Fourier transform based on octave bands is segmented, including: performing A-weighting on the spectrum and segmenting the A-weighting result.

7. The abnormal sound detection method according to claim 1, characterized in that, In response to the existence of a signal segment whose power difference with an adjacent signal segment is greater than a difference threshold, determining the existence of a stable single-frequency abnormal sound includes: in response to the occurrence of a signal segment whose power difference with an adjacent signal segment is greater than a difference threshold more than a number of occurrences, determining that the active noise cancellation system has a stable single-frequency abnormal sound.

8. A control method for an active noise reduction system, characterized in that, include: The abnormal noise detection method according to any one of claims 1-7 is used to detect abnormal noise in the active noise reduction system; In response to the presence of a stable single-frequency abnormal sound in the active noise cancellation system, the active noise cancellation system is restarted.

9. An abnormal noise detection device, characterized in that, The device is applied to an active noise cancellation system, which includes a microphone and a speaker; the abnormal sound detection device is used to implement the abnormal sound detection method according to any one of claims 1-7. The abnormal sound detection device includes: The acquisition module is used to acquire a first noise signal during the operation of the active noise cancellation system; the first noise signal includes the noise signal of the noise source collected by the microphone and / or the control signal of the speaker; The calculation module is used to perform Fourier transform on the first noise signal, segment the spectrum obtained by Fourier transform based on octave bands, and determine the power of each signal segment of the spectrum. The detection module determines the presence of a stable single-frequency abnormal sound in response to the presence of a signal segment whose power difference with the adjacent signal segment is greater than the difference threshold.

10. An active noise cancellation system, comprising a memory, a processor, and a computer program stored in the memory and for running on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 8.

Images