Dual Quadratic Sound Detection System

Abstract

The present application provides a biquad sound detection system, comprising a microphone, an audio detection device, a biquad preprocessing device, a feature extraction device, a feature reduction device, and an event detection device. The microphone receives sound signals in the environment. The audio detection device converts the sound signal into a digital audio signal and outputs a preprocessed signal containing a required frequency band. The biquad preprocessing device preprocesses the digital audio signal and outputs a preprocessed signal containing a required frequency band. The feature extraction device groups the feature values ​​in the preprocessed signal into a plurality of subsets and outputs a feature grouping signal. The feature reduction device sorts the plurality of subsets by importance and outputs a feature sorting signal. The event detection device compares the sound signal according to the importance of the plurality of subsets and outputs an event detection signal having a recognition result.

Description

Technical Field

[0001] The present application provides a sound detection system, and more particularly, a biquadric sound detection system. Background Art

[0002] In real life, there are usually many sound events that occur together and overlap. If the sound events are extracted, they can be used in various applications. For example, fall sound recognition can be applied to emergency rescue, baby crying sound recognition can be applied to baby care, frog croaking sound recognition can be applied to ecological analysis, footstep sound recognition can be applied to home safety, etc.

[0003] Since the sound event detection technology can cover all kinds of sounds, it has great development potential. Among them, the signal preprocessing in the sound event detection is an important module. The filter usually used in the signal preprocessing will use a fixed coefficient or shape filter, which will make the filter unable to be flexibly adjusted. As a result, when the filtering effect is not good, the quality of the sound event detection will be greatly reduced. Summary of the invention

[0004] The embodiment of the present application provides a biquad sound detection system, comprising a microphone, an audio detection device, a biquad preprocessing device, a feature extraction device, a feature reduction device and an event detection device. The microphone receives sound signals in the environment. The audio detection device converts the sound signal through an analog-to-digital converter to output a digital audio signal. The input end of the biquad preprocessing device is connected to the output end of the audio detection device, and the biquad preprocessing device includes a 1 to N order biquad filter. After the digital audio signal passes through the 1 to N order biquad filter, the 1 to N order biquad filter outputs a preprocessed signal containing a required frequency band. The input end of the feature extraction device is connected to the output end of the biquad preprocessing device, and the feature extraction device groups the feature values ​​in the preprocessed signal into multiple subsets and outputs a feature grouping signal. The input end of the feature reduction device is connected to the output end of the feature extraction device, and the feature reduction device sorts the multiple subsets by importance and outputs a feature sorting signal. The input end of the event detection device is connected to the output end of the feature reduction device. The event detection device compares the sound signal according to the importance ranking in the feature ranking signal and outputs an event detection signal with a recognition result.

[0005] It can be seen that the embodiments of the present application are more flexible than the hardware in the prior art and can improve the quality of event detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0007] Figure 1 A schematic diagram of a module of a dual second-order sound detection system provided in an embodiment of the present application.

[0008] Figure 2 A schematic diagram of a module of a biquad filter provided in an embodiment of the present application.

[0009] Figure 3 A schematic diagram of a module for configuring a hierarchical configuration of biquad filters of order 1 to N provided in an embodiment of the present application.

[0010] Figure 4 A schematic diagram of a module of a dual second-order preprocessing device provided in an embodiment of the present application.

[0011] Figure 5 A schematic diagram of a module of a dual second-order preprocessing device provided in another embodiment of the present application.

[0012] Figure 6 A schematic diagram of the flow of a dual second-order sound detection system provided in an embodiment of the present application.

[0013] Figure 7 This is a flow chart of another embodiment of the present application.

[0014] 100: biquad sound detection system; 10: microphone; 20: audio detection device; 30: biquad preprocessing device; 40: feature extraction device; 50: feature reduction device; 60: event detection device. DETAILED DESCRIPTION

[0015] The detailed description and technical content of this application are now described in conjunction with the drawings as follows. Furthermore, the drawings in this application are not necessarily drawn in accordance with the actual proportions for the convenience of explanation, and the drawings and their proportions are not intended to limit the scope of this invention, which is hereby stated in advance.

[0016] The devices, devices, modules and their corresponding functions described in this application can be executed by a single chip or a combination of multiple chips, and the number of chip configurations is not within the scope of this application. In addition, the chip can be, but is not limited to, a combination of devices such as a processor, a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), and a programmable logic device (PLD), which are not limited in this application. The chip can have a shared memory unit, which is used to assist in data storage. In a feasible embodiment, the memory unit can be (but is not limited to) a cache memory, a dynamic random access memory (DRAM), a persistent memory, etc., which can be used as a device or a combination thereof for storing and retrieving data, which is not limited in this application.

[0017] The following is an explanation of one of the embodiments of the present application. Figure 1 , is a module schematic diagram of a dual second-order sound detection system provided in an embodiment of the present application, as shown in the figure:

[0018] See also Figure 1 This embodiment discloses a biquad sound detection system 100, which mainly includes a microphone 10, an audio detection device 20, a biquad preprocessing device 30, a feature extraction device 40, a feature reduction device 50 and an event detection device 60.

[0019] The microphone 10 is mainly used to receive sound signals in the environment. In other embodiments, the microphone 10 may be, for example, a pickup or other device that can be used to receive environmental sound waves. In this application, the sound signal output by the microphone 10 may also be defined as a microphone signal, which is described here first.

[0020] The audio detection device 20 is mainly used to convert the sound signal received by the microphone 10 through an analog-to-digital converter (ADC) to output a digital audio signal. Specifically, the audio detection device 20 includes an analog-to-digital converter (not shown), which converts the received microphone signal into a digital audio signal and outputs it to the biquad preprocessing device 30. In other embodiments, the microphone 10 can be constructed together with the audio detection device 20 as a sound receiving module, so that the functions of receiving the sound signal and converting the digital signal are realized together in the sound receiving module.

[0021] The biquad preprocessing device 30 is mainly used for signal preprocessing, so it can obtain the required sound frequency band through filtering. Here, the signal after signal preprocessing is defined as a preprocessed signal.

[0022] In one embodiment, the biquad preprocessing device 30 includes a 1 to N order biquad filter 32. Specifically, the input end of the 1 to N order biquad filter 321 of the 1 to N order biquad filter 32 is connected to the audio detection device 20. The digital audio signal is filtered by the 1 to N order biquad filter 32. Finally, the N order biquad filter 32N of the 1 to N order biquad filter 32 outputs a preprocessed signal containing a desired audio frequency band to the feature extraction device 40. The biquad filter and the hierarchical configuration of the 1 to N order biquad filter 32 are as follows. Please refer to Figure 2 , Figure 3 : The output of the 1st order biquad filter 321 is connected to the input of the 2nd order biquad filter 322; the output of the 2nd order biquad filter 322 is connected to the input of the 3rd order biquad filter 323, and so on, the output of the N-1st order biquad filter 32N-1 is connected to the input of the Nth order biquad filter 32N. In one embodiment, see Figure 4 The biquad preprocessing device 30 can control the coefficients of the 1st to Nth order biquad filter 32 through a coefficient controller CC, and the output end of the coefficient controller CC is connected to the 1st to Nth order biquad filter 32 to control the coefficients of the 1st to Nth order biquad filter 32.

[0023] In one embodiment, the biquad preprocessing device can automatically update the coefficients according to the input signal and output the preprocessed signal. The hierarchical configuration, biquad filter, and signal input and output of the 1st to Nth order biquad filters in this embodiment are the same as those of the previous embodiment, so they will not be described in detail. Figure 5The biquad preprocessing device 70 sequentially includes a bandwidth detector 72, a coefficient modifier 74 having an input end connected to the bandwidth detector 72, and a 1 to N order biquad filter 76 having an input end connected to the coefficient modifier 74, the other input end of the coefficient modifier 74 is connected to the microphone 10, and a frequency detector (Frequency Detector) (not shown in the figure) is arranged between the coefficient modifier 74 and the microphone 10; wherein, the other input end of the 1 order biquad filter is connected to the audio detection device 20, and the N order biquad filter of the 1 to N order biquad filter 76 outputs a preprocessed signal containing the required audio frequency band to the feature extraction device 40.

[0024] Please refer to Figure 1 , the feature extraction device 40 is mainly used to group the eigenvalues ​​into multiple subsets. The input end of the feature extraction device 40 is connected to the output end of the biquadratic preprocessing device 30, and the preprocessed signal is grouped after passing through an algorithm. The algorithm can be (but not limited to) Mel-Frequency Cepstral Coefficients (MFCCs), Linear Predictive Coding (LPC), Perceptual Linear Prediction Coefficients (PLPC), Linear Prediction Cepstral Coefficients (LPCC), etc. with artificial neural network (ANN) to extract eigenvalues ​​first and then perform classification / grouping operations, which is not limited in this application. The feature extraction device 40 groups the eigenvalues ​​in the preprocessed signal into multiple subsets and then outputs the feature grouping signal to the feature reduction device 50.

[0025] The feature reduction device 50 is mainly used to sort the importance of multiple subsets in the feature grouping signal. The input end of the feature reduction device 50 is connected to the output end of the feature extraction device, and the feature grouping signal is sorted by importance or relevance after algorithm operation. The algorithm can be (but not limited to) minimum redundancy maximum relevance (mRMR), which can be used as an algorithm to compare the relevance of feature values ​​in subsets to reduce the subsequent computing burden, and is not limited in this application. The feature reduction device outputs a feature sorting signal to the event detection device 60 after sorting the multiple subsets by importance.

[0026] The event detection device 60 is mainly used to compare the sound signal according to the importance ranking in the feature ranking signal, and output the recognition result after confirming whether it contains the sound of the expected target. The input end of the event detection device 60 is connected to the output end of the feature reduction device 50, and the corresponding sound is obtained from the sound signal according to the importance ranking of the subset through the algorithm. The algorithm can be but not limited to hidden Markov model (HMM), Gaussian and support vector hybrid (Hybrid GMM-SVM), deep neural network (DNN), Gaussian mixture model (GMM), support vector machine (SVM) and other algorithms that can be applied to data mining / machine learning / artificial intelligence and thus obtain corresponding data, which is not limited in this application. The event detection device 60 compares the corresponding sound through the algorithm and outputs the recognition result, which is defined as the event detection signal.

[0027] The above is a description of a specific embodiment of the hardware architecture of the present application. The operation of the present application will be further described below. Please refer to Figure 6 , is a flow chart of the dual second-order sound detection system 100 of the present application, as shown in the figure: Before the execution of the present application, the dual second-order sound detection system 100 defaults to having a sound event target, and the sound event target is, for example: footsteps, snoring, wind, baby crying, glass breaking, etc., which are not limited in the present application.

[0028] First, the microphone 10 receives a sound signal in the environment and outputs a microphone signal to the audio detection device 20 (step S201 ).

[0029] The microphone signal is outputted as a digital audio signal to the biquad pre-processing device 30 via the analog-to-digital converter in the audio detection device 20 (step S202 ).

[0030] The biquad preprocessing device 30 inputs the received digital audio signal into the 1st-order biquad filter 321, and after passing through the 1st-Nth-order biquad filter 32, the Nth-order biquad filter 32N outputs the preprocessed signal of the required frequency band to the feature extraction device 40 (step S203). In one embodiment, the coefficients of the 1st-Nth-order biquad filter 32 can be factory set or controlled / adjusted via a coefficient controller, which is not limited in the present application.

[0031] Please refer to the coefficient formula of the 1st to Nth order biquad filter 32 in this embodiment. Figure 2 , Figure 3, the biquad filters 321-32N of the 1st to Nth order biquad filters 32 filter the digital audio signal according to the following formula: y[n]=b0*x[n]+b1*x[n-1]+b2*x[n-2]-a1*y[n-1]-a2*y[n-2];

[0032] Among them, x[n], x[n-1], x[n-2] are the digital audio signals received at the nth order, n-1th order and n-2th order time points respectively, y[n], y[n-1], y[n-2] are the preprocessed signals output at the nth order, n-1th order and n-2th order time points respectively, and b0, b1, b2, a1, a2 are the coefficients of the biquadratic filter.

[0033] In one embodiment, if the default sound event target is a crying baby, and in the case of a 1st to 4th order biquad filter, the coefficients of the 1st to 4th order biquad filters are set as follows:

[0034] 1st order: coefficients b0=0.5671, b1=1.1342, b2=0.5671, -a1=0.9396, a2=0.3288.

[0035] 2nd order: coefficients b0=0.5671, b1=1.1342, b2=0.5671, -a1=0.9396, a2=0.3288.

[0036] 3rd order: coefficients b0=0.8940, b1=-1.7879, b2=0.8940, -a1=-1.7768, a2=0.7990.

[0037] 4th order: coefficients b0=0.8940, b1=-1.7879, b2=0.8940, -a1=-1.7768, a2=0.7990.

[0038] According to the coefficient setting, the digital audio signal output by the 4th-order biquad filter is mainly in the sound frequency band of 0.2kHz to 1.2kHz.

[0039] In another embodiment, if the default sound event target is glass breaking sound, and in the case of a 1st to 4th order biquad filter, the coefficients of the 1st to 4th order biquad filters are set as follows:

[0040] 1st order: coefficients b0=0.0976, b1=0.1952, b2=0.0976, -a1=-0.9427, a2=0.3332.

[0041] 2nd order: coefficients b0=0.0976, b1=0.1952, b2=0.0976, -a1=-0.9427, a2=0.3332.

[0042] 3rd order: coefficients b0=0.9932, b1=-1.9865, b2=0.9932, -a1=-1.9865, a2=0.9866.

[0043] 4th order: coefficients b0=0.9932, b1=-1.9865, b2=0.9932, -a1=-1.9865, a2=0.9866.

[0044] According to the coefficient setting, the digital audio signal output by the 4th-order biquad filter is mainly in the sound frequency band of 0.01kHz to 0.4kHz.

[0045] Next, the feature extraction device 40 groups the feature values ​​in the pre-processed signal into a plurality of subsets and outputs the feature clustering signal to the feature reduction device 50 (step S204 ).

[0046] In one embodiment, the operation of eigenvalue clustering uses Mel-frequency cepstral coefficients for analysis. The received digital audio signal is first pre-emphasized (also called pre-enhancement) and frame blocked and windowed, and then the corresponding spectrum is obtained through Fourier transform (FFT). The obtained spectrum is then passed through a Mel filter group to obtain a Mel spectrum. Cepstral analysis is performed on the Mel spectrum (logarithm, inverse transformation), and the Mel-frequency cepstral coefficients MFCCs are obtained. After that, they are clustered into multiple subsets through an artificial neural network according to the obtained eigenvalues.

[0047] The feature reduction device 50 sorts the subsets of the feature clustering signals by importance and then outputs a feature sorting signal to the event detection device 60 (step S205 ).

[0048] In one embodiment, the importance ranking operation uses minimum redundancy maximum correlation to maximize the correlation between features and classification variables and minimize the correlation between features for a subset of feature clustering signals, and then uses mutual information (MI) as a measurement standard to obtain the importance ranking of the subset.

[0049] Finally, the event detection device 60 receives the feature ranking signal and compares the feature values ​​from the sound signal according to the importance ranking. After the comparison is completed and the recognition result of the sound corresponding to the sound event target is obtained, the event detection device 60 outputs an event detection signal including the recognition result (step S206).

[0050] In one embodiment, the comparison algorithm of the present application uses a hidden Markov model as an acoustic model for modeling, and performs training analysis through a forward algorithm, a backward search algorithm, or a Viterbi algorithm. After the importance ranking is compared with the sound event target through the hidden Markov model, it is compared from the sound signal whether it contains a sound corresponding to the sound event target and an event detection signal containing the identification result is output.

[0051] In another embodiment, when the biquad preprocessing device 70 of the present application has a bandwidth detector 72, a coefficient modifier 74, and a 1 to N order biquad filter 76, the step S203 is replaced by steps S2031-2034, see Figure 7 , is a flow chart of another embodiment of the present application, wherein the flow chart does not disclose steps S201, S205, and S206; in this embodiment, the steps and the same formulas that have been described are not repeated.

[0052] Continuing from step S202 , the biquad pre-processing device 70 inputs the received digital audio signal into the 1st to Nth order biquad filter 76 and the bandwidth detector 72 (step S2031 ).

[0053] The bandwidth detector 72 detects the bandwidth of the digital audio signal and calculates and outputs a bandwidth signal with the same bandwidth as the center frequency of the digital audio signal to the coefficient modifier 74 (step S2032). The center frequency is obtained by the bandwidth detector 72 through the digital audio signal according to the following formula:

[0054]

[0055] k=0,……,M-1;

[0056] Wherein, x[n] is the digital audio signal outputted from the audio detection device 20 to the bandwidth detector 72 at stage n, and f k is the center frequency output by the bandwidth detector 72, f k There are M outputs in total, where M is the preset number of outputs.

[0057] The coefficient modifier 74 calculates and outputs a coefficient modification signal to the 1st to Nth order biquad filter 76 according to the bandwidth signal and the frequency of the received microphone signal (step S2033). Figure 2 , the coefficient modifier 74 modifies the coefficients of the biquad filters in the 1st to Nth order biquad filters 76 according to the following formula:

[0058]

[0059]

[0060]

[0061]

[0062]

[0063] Among them, w0 is the center angular frequency value, α is the natural frequency parameter, and b0, b1, b2, a1, and a2 are the coefficients of the biquad filter.

[0064] The central angular frequency value and the natural frequency parameter are obtained by the coefficient corrector 34 according to the following formula:

[0065]

[0066]

[0067] Among them, f k is the center frequency input by the bandwidth detector 72, F s is the frequency of the microphone signal input by the microphone 10, Q is the default quality parameter, w0 is the center angular frequency value, and α is the natural frequency parameter. The quality parameter mentioned above is determined based on the quality factor; the natural frequency parameter is determined based on the natural frequency factor.

[0068] Next, the 1-N order biquad filter 76 modifies the coefficients according to the received coefficient modification signal, and pre-processes (filters) the digital audio signal received in the next stage according to the modified coefficients (x[n-1], x[n-2] are regarded as known parameters) and then outputs the pre-processed signal to the feature extraction device 40 (step S2034). This step will be followed by step S204, which is described here.

[0069] In summary, compared to the prior art, the hardware of the present application is more flexible and can improve the quality of event detection.

[0070] The present application has been described in detail above. However, what is described above is only a preferred embodiment of the present application and should not be used to limit the scope of implementation of the present application. That is, all equivalent changes and modifications made according to the claims of the present application should still fall within the scope of the patent coverage of the present application.

Claims

1. A dual quadratic sound detection system, characterized in that: include: Before the biquadratic sound detection system is executed, there is a default sound event target, and the sound event target includes: footsteps, snoring, wind, baby crying and glass breaking; Microphone, receiving sound signals from the environment; An audio detection device converts the sound signal via an analog-to-digital converter to output a digital audio signal; A biquad preprocessing device, the input end of which is connected to the output end of the audio detection device, the biquad preprocessing device comprises a 1st to Nth order biquad filter, and after the digital audio signal passes through the 1st to Nth order biquad filter, the 1st to Nth order biquad filter outputs a preprocessed signal containing a required frequency band; A feature extraction device, whose input end is connected to the output end of the biquadratic preprocessing device, and the feature extraction device groups the feature values ​​in the preprocessed signal into multiple subsets and then outputs a feature grouping signal; a feature reduction device, whose input end is connected to the output end of the feature extraction device, wherein the feature reduction device sorts the plurality of subsets by importance and outputs a feature sorting signal; and an event detection device, the input end of which is connected to the output end of the feature reduction device, the event detection device compares the sound signal according to the importance ranking in the feature ranking signal and outputs an event detection signal with a recognition result; The biquad filter filters the digital audio signal according to the following formula: y[n]=b0*x[n]+b1*x[n-1]+b2*x[n-2]–a1*y[n-1]-a2*y[n-2]; Wherein, x[n], x[n-1], x[n-2] are the digital audio signals received at the nth order, n-1th order and n-2th order time points, respectively; y[n], y[n-1], y[n-2] are the preprocessed signals output at the nth order, n-1th order and n-2th order time points, respectively; b0, b1, b2, a1, a2 are the coefficients of the biquad filter; If the sound event target is a crying baby, and in the case of a 1st to 4th order biquad filter, the coefficients of the 1st to 4th order biquad filters are set as follows: 1st order: coefficients b0 = 0.5671, b1 = 1.1342, b2 = 0.5671, -a1 = 0.9396, a2＝0.3288； 2nd order: coefficients b0 = 0.5671, b1 = 1.1342, b2 = 0.5671, -a1 = 0.9396, a2＝0.3288； 3rd order: coefficients b0 = 0.8940, b1 = -1.7879, b2 = 0.8940, -a1 = -1.7768, a2＝0.7990； 4th order: coefficients b0 = 0.8940, b1 = -1.7879, b2 = 0.8940, -a1 = -1.7768, a2 = 0.7990; If the sound event target is glass breaking sound and the filter is a 1st to 4th order biquad filter, the coefficients of the 1st to 4th order biquad filters are set as follows: 1st order: coefficients b0 = 0.0976, b1 = 0.1952, b2 = 0.0976, -a1 = -0.9427, a2＝0.3332； 2nd order: coefficients b0 = 0.0976, b1 = 0.1952, b2 = 0.0976, -a1 = -0.9427, a2＝0.3332； 3rd order: coefficients b0 = 0.9932, b1 = -1.9865, b2 = 0.9932, -a1 = -1.9865, a2＝0.9866； 4th order: coefficients b0=0.9932, b1=-1.9865, b2=0.9932, -a1=-1.9865, a2=0.9866.

2. The biquad sound detection system as claimed in claim 1, wherein: The input end of the 1st-order biquad filter is connected to the audio detection device to receive the digital audio signal. After the digital audio signal passes through the 1st to Nth-order biquad filters, the Nth-order biquad filter outputs the preprocessed signal.

3. The biquad sound detection system as claimed in claim 1, wherein: The biquad preprocessing device comprises a coefficient controller, wherein an output terminal of the coefficient controller is connected to the 1st to Nth order biquad filters to control coefficients of the 1st to Nth order biquad filters.

4. The biquad sound detection system as claimed in claim 1, wherein: The biquad preprocessing device comprises a bandwidth detector, a coefficient modifier having an input end connected to an output end of the bandwidth detector; An input end of the 1st to Nth order biquad filter is connected to the coefficient modifier, and another input end of the coefficient modifier is connected to the output end of the microphone; The bandwidth detector detects the bandwidth of the digital audio signal and outputs a bandwidth signal with the same bandwidth as the center frequency of the digital audio signal; the coefficient modifier calculates and outputs a coefficient correction signal for correcting the coefficients of the 1st to Nth order biquad filter according to the bandwidth signal and the frequency of the sound signal; the 1st to Nth order biquad filter updates the coefficient according to the coefficient correction signal, and pre-processes the digital audio signal received in the next stage according to the corrected coefficient and outputs the pre-processed signal.

5. The biquad sound detection system as claimed in claim 4, wherein: A frequency detector is provided between the coefficient modifier and the microphone.

6. The biquad sound detection system as claimed in claim 4, wherein: The coefficient modifier modifies the coefficients of the biquad filter according to the following formula: Wherein, w0 is the center angular frequency value, α is the natural frequency parameter, and b0, b1, b2, a1, and a2 are the coefficients of the biquad filter.

7. The biquad sound detection system as claimed in claim 6, wherein: The central angular frequency value and the natural frequency parameter are obtained by the coefficient corrector according to the following formula: Among them, f k is the center frequency input by the bandwidth detector, F s is the frequency input by the microphone, Q is the default quality parameter, w0 is the center angular frequency value, and α is the natural frequency parameter.

8. The biquad sound detection system as claimed in claim 7, wherein: The quality parameter is determined based on a quality factor, and the natural frequency parameter is determined based on a natural frequency factor.

9. The biquad sound detection system as claimed in claim 7, wherein: The center frequency is obtained by the bandwidth detector via the digital audio signal according to the following formula: k=0,……,M-1; Wherein, x[n] is the digital audio signal output by the audio detection device at stage n, and f k There are M outputs in total, where M is the preset number of outputs.

Images