A sound detection method

Abstract

The application discloses a sound detection method and relates to the technical field of measuring electric variables; the method comprises the following steps: S1, audio signals are segmented by using a sliding wake-up time window, and multi-channel root mean square values are calculated; when the multi-channel root mean square values are greater than a preset wake-up threshold, it is determined that wake-up is triggered; S2, the audio signals are segmented by using a sliding interaction time window to obtain audio segments of each channel; each audio segment is further divided to obtain multiple equal-length sub-segments; single-channel root mean square values are calculated; when the single-channel root mean square values are greater than an activation threshold, it is determined that there is an effective signal in the sub-segments; the wake-up detection in step S1 and the formation of a lightweight signal processing method in step S2 take into account real-time performance and deployment cost, and the efficiency of recognizing touch sequences based on a microphone array is improved; in step S2, the multiple equal-length sub-segments are further divided, the single-channel root mean square values of the sub-segments are compared with the activation threshold, and the recognition accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of electrical variable measurement technology, and more particularly to a sound detection method. Background Technology

[0002] Currently, human-computer interaction methods using multi-channel microphone arrays typically involve adding pressure or touch sensors near the microphones to identify the touch sequence, thereby enabling human-computer interaction and enabling applications in multi-frequency smart speaker devices. This application argues that existing solutions suffer from the following technical problems in identifying the touch sequence.

[0003] First, it leads to adjustments in hardware structure, changes in molds, and additional component costs, thereby reducing the product's economic viability and market competitiveness.

[0004] Second, embedded devices have limited computing resources and it is difficult to acquire training data on a large scale.

[0005] Third, although deep learning methods perform well in complex acoustic classification tasks, they have high requirements for computing resources and data scale, making them unsuitable for this technical scenario.

[0006] Therefore, the low efficiency of microphone array-based touch sequence recognition has become a pressing technical problem that needs to be solved. Summary of the Invention

[0007] This invention provides a sound detection method that solves the technical problem of low efficiency in identifying touch sequence based on microphone arrays.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] A sound detection method includes the following steps:

[0010] Step S1: Obtain the audio signal of each channel of the microphone array, segment the audio signal using a sliding wake-up time window, calculate the root mean square value of the multi-channel signal within the wake-up time window, and determine that wake-up is triggered when the root mean square value of the multi-channel signal is greater than the preset wake-up threshold, and proceed to step S2; when the root mean square value of the multi-channel signal is less than or equal to the preset wake-up threshold, determine that wake-up is not triggered.

[0011] Step S2: Obtain the audio signal of each channel of the microphone array. Use a sliding interactive time window to segment the audio signal to obtain audio segments for each channel. Further divide each audio segment into multiple equal-length sub-segments. Calculate the root mean square (RMS) value of each sub-segment of each channel's audio segment. When the RMS value is greater than the activation threshold, a valid signal is determined to exist in the sub-segment. Based on the valid signals in each channel's sub-segments, establish a time index t for the first detection of sound in the corresponding channel.c It also enables time alignment of all channels under interactive behavior, and identifies the timing result of the microphone being touched in the microphone array based on the time index of the first sound detection of all channels.

[0012] A further technical solution is as follows: In step S1, the microphone array is a four-channel microphone array, and the audio signals of the four channels are obtained based on the four-channel microphone array. The root mean square value of the multi-channel is the root mean square value of the four channels. The controller calculates the root mean square value of the four channels within the wake-up time window according to formula (1).

[0013] (1)

[0014] In equation (1), RMS4 is the root mean square value of the four channels of a sub-segment of a four-channel audio segment. Indicates the audio channel number; This indicates the sampling point number within the wake-up time window; N represents the number of sampling points per channel. This represents the nth sample value of the c-th channel;

[0015] In step S2, the obtained timing result is the interaction category, which includes single-point touch with the first microphone, single-point touch with the second microphone, single-point touch with the third microphone, single-point touch with the fourth microphone, swipe left, swipe right, and invalid trigger.

[0016] A further technical solution is that, in step S1, the wake-up time window length is 0.1s and the wake-up sliding step size is 0.1s.

[0017] A further technical solution is that, in step S2, after obtaining the timing result, the corresponding interactive instruction is obtained based on the timing result.

[0018] A further technical solution is that, in step S2, the root mean square value of each sub-segment of each audio channel segment is calculated according to equation (2).

[0019] (2)

[0020] In equation (2), RMS1 is the root mean square value of a single channel audio segment; K is the number of sampling points contained in each sub-segment; and k is the kth sampling point. This is the kth sample value of a mono channel.

[0021] A further technical solution is as follows: In step S2, the technical conditions for time alignment are shown in equation (3).

[0022] (3)

[0023] Equation (3) indicates that at the same time, at least one of the first sub-segments of the four channels contains a valid signal, when t c When =None, t c Consider it as +∞. Indicates the channel number.

[0024] A further technical solution is as follows: In step S2, the interaction time window length is 0.2s, the interaction sliding step length is 0.02s, and an audio segment is divided into num sub-segments of equal length, where num=10.

[0025] The beneficial effects of adopting the above technical solution are as follows:

[0026] A sound detection method includes step S1, which segments the audio signal using a sliding wake-up time window and calculates the multi-channel root mean square (RMS) value within the wake-up time window. When the multi-channel RMS value is greater than a preset wake-up threshold, wake-up is triggered. Step S2, which segments the audio signal using a sliding interaction time window to obtain audio segments for each channel, further divides each audio segment into multiple equal-length sub-segments, and calculates the mono-channel RMS value for each sub-segment of each channel audio segment. When the mono-channel RMS value is greater than an activation threshold, a valid signal is determined to exist in the sub-segment. Step S1 wake-up detection and step S2 form a lightweight signal processing method that balances real-time performance and deployment cost, improving the efficiency of microphone array-based touch sequence recognition. In step S2, each audio segment is further divided into multiple equal-length sub-segments, and the mono-channel RMS value of each sub-segment is obtained and compared with the activation threshold for filtering, thus improving recognition accuracy. Attached Figure Description

[0027] Figure 1 This is a principle block diagram of Embodiment 1 of the present invention;

[0028] Figure 2 This is a diagram showing the distribution of a four-channel microphone array;

[0029] Figure 3 This is a flowchart of the wake-up detection phase;

[0030] Figure 4 This is a flowchart of the interactive recognition phase;

[0031] Figure 5a It is the original audio time-domain graph of the wake-up detection phase;

[0032] Figure 5b This is a screenshot of the tapping test results during the wake-up detection phase;

[0033] Figure 6a It is the original audio time-domain graph of the first microphone during the single-point touch recognition stage;

[0034] Figure 6b This is the original audio time-domain plot of the second microphone during the single-point touch interaction recognition stage;

[0035] Figure 6c This is the original audio time-domain plot of the third microphone during the single-point touch interaction recognition stage;

[0036] Figure 6d This is the original audio time-domain plot of the fourth microphone during the single-point touch interaction recognition stage;

[0037] Figure 6e This is the original audio temporal plot of the left swipe during the interactive recognition phase;

[0038] Figure 6f This is the original audio temporal plot of the right swipe during the interactive recognition phase;

[0039] Figure 7 This is a data graph of right-swipe classification during the interactive recognition stage.

[0040] The components are: 1. First microphone, 2. Second microphone, 3. Third microphone, 4. Fourth microphone, and 5. Cabinet. Detailed Implementation

[0041] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this application or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0042] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below. Example 1

[0043] This invention discloses a sound detection method based on a host computer and a sound detection unit, including a step of recognizing touch.

[0044] like Figure 1 As shown, the sound detection unit includes a controller and a multi-channel microphone array. The multi-channel microphone array is a four-channel microphone array. The multi-channel microphone array is electrically connected to the controller, and the controller is electrically connected to the host computer and communicates bidirectionally. The host computer, the controller, and the multi-channel microphone array form a hardware structure.

[0045] The controller model is SIM8950-A. The multi-channel microphone array is a four-channel microphone array formed by four microphones. The four-channel microphone array includes a first microphone 1, a second microphone 2, a third microphone 3, and a fourth microphone 4. The four microphones correspond to four independent channels. The first microphone 1 corresponds to the first channel channel 1, the second microphone 2 corresponds to the second channel channel 2, the third microphone 3 corresponds to the third channel channel 3, and the fourth microphone 4 corresponds to the fourth channel channel 4. The four microphones are used to collect acoustic signals generated by the user at different positions.

[0046] like Figure 2 As shown, the positions of the four microphones form a square.

[0047] This embodiment uses hardware such as a host computer, controller, and multi-channel microphone array to detect sound and identify the touch sequence. This embodiment can be understood as the most basic technical solution. When specifically applied to a multi-frequency smart speaker device, four microphones are fixedly connected to the top of the speaker. The speaker includes a cabinet 5, a controller, and a playback component. The controller and playback component are fixedly connected to the cabinet 5. The four microphones are evenly distributed on the top surface of the speaker cabinet 5, and the lines connecting the positions of the four microphones form a square. Each microphone corresponds to one audio channel.

[0048] The touch recognition process employs a pure software algorithm to achieve interactive recognition based on acoustic signals without adding any additional hardware, as explained below.

[0049] To enhance the intuitiveness and control experience of user interaction, this application proposes an interaction mechanism based on haptic-acoustic feedback: users can wake up the system and trigger corresponding functions by tapping or swiping at the microphone. In interaction mode, the system supports the following six types of operation commands.

[0050] Swipe left: The gesture moves from the second microphone 2 to the first microphone 1, or from the third microphone 3 to the fourth microphone 4, or from the second microphone 2 to the fourth microphone 4, or from the third microphone 3 to the first microphone 1, to switch to the previous song.

[0051] Right swipe: Swipe from the first microphone 1 to the second microphone 2, or from the fourth microphone 4 to the third microphone 3, or from the first microphone 1 to the third microphone 3, or from the fourth microphone 4 to the second microphone 2 to switch to the next song.

[0052] Single-point touch: Trigger the first microphone 1, the second microphone 2, the third microphone 3 or the fourth microphone 4 respectively. Each microphone corresponds to a predefined function, for a total of four predefined functions.

[0053] To address the practical constraints of limited computing resources and the difficulty in acquiring large-scale training data for embedded devices, this application compares deep learning methods with traditional signal processing methods. While deep learning methods perform exceptionally well in complex acoustic classification tasks, their high requirements for computing resources and data scale make them unsuitable for this scenario. Therefore, this application employs a lightweight signal processing method to balance real-time performance with deployment costs.

[0054] Specifically, the touch recognition steps include two core stages: the first stage is the wake-up detection stage, and the second stage is the interaction recognition stage.

[0055] Step S1: Wake-up detection phase.

[0056] like Figure 3 As shown, the wake-up detection steps include continuously monitoring the audio signals of the four channels with four microphones in real time, segmenting the audio signals using a sliding wake-up time window, and calculating the root mean square (RMS) values ​​of the four channels within the wake-up time window according to equation (1). When the RMS value of the four channels is greater than the preset wake-up threshold, it is determined that wake-up has been triggered and the interactive recognition step is entered; when the RMS value of the four channels is less than or equal to the preset wake-up threshold, it is determined that wake-up has not been triggered. The length of the wake-up time window is a set value, namely the wake-up time window length frame_len_wake, and the sliding step size of the wake-up time window is a set value, namely the wake-up sliding step size hop_size_wake. Details are as follows.

[0057] By analyzing the energy characteristics of the input multi-channel audio signal, it can be determined whether there is a significant tapping or triggering action, thereby deciding whether the system should enter interactive mode.

[0058] Since the microphone continuously monitors in real time after the device is powered on, various interfering noises are inevitable in the environment, such as human voices and background noise. In contrast, when a user taps near the microphone, the sound source is closer to the microphone and has higher energy characteristics. Therefore, the energy distribution differences of multi-channel audio signals can be used to effectively detect wake-up actions.

[0059] This application uses the overall root mean square value of the four microphone channel signals as the energy characterization index. The root mean square value is abbreviated as RMS. The overall root mean square value of the four microphone channel signals is the four-channel root mean square value, and the calculation formula is as follows.

[0060] (1)

[0061] In equation (1), RMS4 is the root mean square value of the four channels of a sub-segment of a four-channel audio segment. Indicates the audio channel number; Indicates the sampling point number; N represents the number of sampling points per channel; This represents the nth sample value of the c-th channel.

[0062] When the calculated root mean square value of the four channels is greater than the preset wake-up threshold_wake, it is determined that wake-up is triggered and the detection result is 1; when the calculated root mean square value of the four channels is less than or equal to the preset wake-up threshold_wake, it is determined that wake-up is not triggered and the detection result is 0.

[0063] Considering the need for real-time processing, a sliding time window is used to segment and calculate the audio signal. The wake-up time window length is set to frame_len_wake, and the wake-up sliding step size is set to hop_size_wake.

[0064] In this experiment, the controller and multi-channel microphone array were temporarily fixed to the top of the speaker. The experimental data was obtained by having the user tap the microphone array area of ​​the speaker, and then using wake-up detection to test and continuously adjust the parameters.

[0065] The wake-up time window length `frame_len_wake` is set to 0.1s, and the wake-up sliding step size `hop_size_wake` is also set to 0.1s. With a sampling rate of 16kHz, the corresponding number of sampling points per frame is N=1600. That is, the root mean square value of the four channels is calculated every 0.1s, and a wake-up determination is made. The preset wake-up threshold `threshold_wake` is 0.1.

[0066] The experimental results of the wake-up detection phase are explained below.

[0067] like Figure 5a As shown, this is a time-domain graph of the original audio during the wake-up detection phase. The horizontal axis represents time in seconds, and the vertical axis represents amplitude.

[0068] like Figure 5b The graph shown represents the tapping detection results during the wake-up detection phase. The horizontal axis represents time in seconds, and the vertical axis represents the detection result, with 1 indicating a tap and 0 indicating no tapping. A detection value is output every 0.1 seconds, where 0 indicates no wake-up and 1 indicates wake-up.

[0069] As can be seen, despite some interference from speech or background noise, the wake-up signal, namely the sound signal generated by tapping the microphone array of the speaker, can still be accurately identified based on the preset wake-up threshold.

[0070] Step S2: Interactive recognition stage.

[0071] Once the wake-up detection phase is activated, it automatically enters the interactive recognition mode.

[0072] like Figure 4 As shown, the interactive recognition steps include: continuously monitoring the audio signals of the four channels in real time using four microphones; segmenting the audio signals using a sliding interactive time window to obtain audio segments for each channel; further dividing each audio segment into num equal-length sub-segments; and calculating the root mean square value of each sub-segment of each channel audio segment according to equation (2). When the root mean square value of the single channel is greater than the activation threshold, it is determined that there is a valid signal in the sub-segment of that channel at that moment. The length of the interactive time window is a set value, namely the interactive time window length frame_len_interacte, and the sliding step size of the interactive time window is a set value, namely the interactive sliding step size hop_size_interacte. Based on the valid signal in each channel sub-segment, a time index t is established to obtain the time when the corresponding channel first detects sound. c This process enables time alignment of the four audio channels during interactive behavior, classifies the interaction category based on the time index of the first sound detection on each of the four channels, and obtains the corresponding interaction command based on the interaction category. Details are as follows.

[0073] In interactive mode, the system uses the temporal sequence and energy distribution characteristics of the four-channel audio signals to identify the user's touch or swipe actions, thereby recognizing six types of interactive commands.

[0074] like Figure 6a As shown, this is the original audio time-domain graph of the first microphone during the single-point touch interaction recognition stage. The horizontal axis represents time in seconds, and the vertical axis represents amplitude.

[0075] like Figure 6b As shown, this is the original audio time-domain graph of the second microphone during the single-point touch interaction recognition stage. The horizontal axis represents time in seconds, and the vertical axis represents amplitude.

[0076] like Figure 6c As shown, this is the original audio time-domain plot of the third microphone during the single-point touch interaction recognition stage. The horizontal axis represents time in seconds, and the vertical axis represents amplitude.

[0077] like Figure 6d As shown, this is the original audio time-domain plot of the fourth microphone during the single-point touch interaction recognition stage. The horizontal axis represents time in seconds, and the vertical axis represents amplitude.

[0078] like Figure 6e As shown, this is the time-domain plot of the original audio during the left swipe in the interactive recognition stage. The horizontal axis represents time in seconds, and the vertical axis represents amplitude.

[0079] like Figure 6f As shown, this is the time-domain plot of the original audio from the right swipe during the interactive recognition stage. The horizontal axis represents time in seconds, and the vertical axis represents amplitude.

[0080] Next based on Figure 6f The interaction time window, audio segment, and sub-segment are explained below.

[0081] like Figure 7 The diagram shows the data for right-slide classification during the interactive recognition phase. To achieve temporal analysis, this application uses a sliding time window to segment the audio signal. The interaction time window length is set to `frame_len_interacte`, and the interaction sliding step size is `hop_size_interacte`. Each audio segment of length `frame_len_interacte` corresponds to one classification decision. If the current segment does not belong to any predefined category, the interaction sliding step size `hop_size_interacte` is used to slide forward to the next audio segment, continuing the judgment until a valid interaction category is identified or the processing ends.

[0082] Step S201: Timing modeling and trigger time definition.

[0083] Considering the obvious temporal sequence of user interactions, this application further divides the audio segment of each channel within each interaction time window into num equal-length sub-segments. The sub-segments obtained by dividing an audio segment into one channel are denoted as:

[0084]

[0085] The initial trigger time position is determined by detecting the energy response of each channel in different sub-segments.

[0086] The root mean square value of a single channel is calculated for each sub-segment of each audio segment to determine whether there is a valid signal within that time segment. Assuming that each sub-segment contains K sampling points, the root mean square value of a single channel is calculated according to equation (2).

[0087] (2)

[0088] In equation (2), RMS1 is the root mean square value of a single channel audio segment; K is the number of sampling points contained in each sub-segment; and k is the kth sampling point. This is the kth sample value of a mono channel.

[0089] When the root mean square value of a certain channel in a certain sub-segment is greater than the activation threshold When the signal is detected, it is determined that there is a valid signal in the sub-segment of that channel at that moment.

[0090] A time index for the first detection of sound in the corresponding channel is established based on the effective signal in each channel sub-segment, as defined below.

[0091] : t1 represents the time index at which the first microphone 1, corresponding to the first channel 1, first detects sound; t2 represents the time index at which the second microphone 2, corresponding to the second channel 2, first detects sound; t3 represents the time index at which the third microphone 3, corresponding to the third channel 3, first detects sound; and t4 represents the time index at which the fourth microphone 4, corresponding to the fourth channel 4, first detects sound. When no valid signal is detected in a channel, it is marked as None.

[0092] like Figure 7 As shown, num=10, and within the interactive time window, each audio segment is divided into 10 sub-segments, t4=1, t1=2, t3=4, t2=5.

[0093] To ensure the time alignment of interactive behaviors, the technical conditions for time alignment of interactive behaviors are shown in Equation (3).

[0094] (3)

[0095] Equation (3) indicates that at the same time, at least one of the first sub-segments of the four channels contains a valid signal, when t c When =None, t c Consider it as +∞. Indicates the channel number.

[0096] When at least one of the first sub-segments of the four channels has a valid signal at the same time, it represents the time alignment of the interaction behavior so that the current sub-segment can participate in the next classification step. Otherwise, the interaction time window is defined as an invalid trigger, i.e., category 0, until a sub-segment that satisfies equation (3) appears.

[0097] Step S202: Spatial distribution and orientation determination.

[0098] like Figure 1 and Figure 2 As shown, based on the spatial distribution of the microphones, the first microphone 1 and the fourth microphone 4 are located on the left side of the device, while the second microphone 2 and the third microphone 3 are located on the right side of the device.

[0099] By combining the chronological order, the sliding direction can be determined.

[0100] If the second channel or third channel of the second microphone 2 or the third microphone 3 on the right first obtains a valid signal, and the first channel or fourth channel of the first microphone 1 or the fourth microphone 4 on the left subsequently obtains a valid signal, then it is determined to be a left slide.

[0101] If the first channel or fourth channel of the first microphone 1 or the fourth microphone 4 on the left obtains a valid signal first, and the second channel or third channel of the second microphone 2 or the third microphone 3 on the right subsequently obtains a valid signal, then it is determined as a right slide.

[0102] If only one of the first microphone 1, the second microphone 2, the third microphone 3, or the fourth microphone 4 receives a valid signal, it is determined to be a single-point touch.

[0103] Step S203: Define the interaction category.

[0104] The system defines six types of interactive commands, as follows.

[0105] First microphone 1 single-point touch: Category 1, ,and .

[0106] Second microphone 2 single-point touch: Category 2, ,and .

[0107] Third microphone 3-point touch: Third category, ,and .

[0108] Fourth microphone 4-point touch: Category 4 ,and .

[0109] Swipe left: Category 5 or And t1≥2 or t4≥2, and t1 or .

[0110] Swipe right: Category 6 or And t2≥2 or t3≥2, and t2 or .

[0111] All other cases are defined as invalid triggers, category 0.

[0112] Experimental results show that the system performs best under the following parameter configurations: interaction time window length: frame_len_interacte=0.2s, interaction sliding step size: hop_size_interacte=0.02s, sampling rate: 16kHz.

[0113] Furthermore, the optimal temporal resolution and recognition performance can be obtained by dividing each interaction time window into num=10 sub-segments.

[0114] Experimental results show that this method has good real-time performance and high recognition accuracy on low computing power platforms, and can effectively replace additional sensors to achieve a low-cost and highly reliable human-computer interaction solution. Example 2

[0115] The present invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the corresponding steps in the sound detection method described in Embodiment 1.

[0116] The computer program includes a wake-up detection module and an interaction recognition module, as described below.

[0117] The wake-up module is used to obtain audio signals from four channels. It uses a sliding wake-up time window to segment the audio signals and calculates the root mean square value of the four channels within the wake-up time window according to formula (1). When the root mean square value of the four channels is greater than the preset wake-up threshold, it is determined to trigger wake-up and enter the interactive recognition step; when the root mean square value of the four channels is less than or equal to the preset wake-up threshold, it is determined not to trigger wake-up.

[0118] The interactive recognition module is used to obtain audio signals from four channels. A sliding interactive time window is used to segment the audio signals to obtain audio segments for each channel. Each audio segment is further divided into num equal-length sub-segments. The root mean square value (RMS) of each sub-segment of each channel's audio segment is calculated according to equation (2). When the RMS value is greater than the activation threshold, it is determined that there is a valid signal in the sub-segment of that channel at that moment. Based on the valid signal in each channel's sub-segment, a time index t is established to obtain the first sound detected in the corresponding channel. c It also enables time alignment of the four channels under interactive behavior, classifies the interaction category based on the time index of the first sound detection of the four channels, and obtains the corresponding interaction command based on the interaction category.

[0119] Compared to Embodiment 1, the multi-channel microphone array is a dual-channel microphone array, which can also achieve interactive recognition. The similarities will not be repeated here.

[0120] Compared to Embodiment 1, the multi-channel microphone array is an eight-channel microphone array, that is, a 2x4 microphone array can also achieve interactive recognition. There are two microphones horizontally and four rows of microphones vertically. The similarities will not be repeated.

[0121] Compared to Example 1, theoretically the multi-channel microphone array can also be a multi-channel microphone array. In the interactive recognition step, the timing result of the microphone being touched is obtained according to the time index of the first sound detection of the multi-channel microphone. The timing result is the touch trajectory. The similarities will not be repeated.

Claims

1. A sound detection method, characterized in that: Includes the following steps, Step S1: Obtain the audio signal of each channel of the microphone array, segment the audio signal using a sliding wake-up time window, calculate the root mean square value of the multi-channel signal within the wake-up time window, and determine that wake-up is triggered when the root mean square value of the multi-channel signal is greater than the preset wake-up threshold, and proceed to step S2; when the root mean square value of the multi-channel signal is less than or equal to the preset wake-up threshold, determine that wake-up is not triggered. In step S1, the microphone array is a four-channel microphone array. The audio signals of the four channels are obtained based on the four-channel microphone array. The root mean square value of the multi-channel is the root mean square value of the four channels. The controller calculates the root mean square value of the four channels within the wake-up time window according to equation (1). （1） In equation (1), RMS4 is the root mean square value of the four channels of a sub-segment of a four-channel audio segment. Indicates the audio channel number; Indicates the sampling point number; N represents the number of sampling points within the wake-up time window of each channel. This represents the nth sample value of the c-th channel; Step S2: Obtain the audio signal of each channel of the microphone array. Use a sliding interactive time window to segment the audio signal to obtain audio segments for each channel. Further divide each audio segment into multiple equal-length sub-segments. Calculate the root mean square (RMS) value of each sub-segment of each channel's audio segment. When the RMS value is greater than the activation threshold, a valid signal is determined to exist in the sub-segment. Based on the valid signals in each channel's sub-segments, establish a time index t for the first detection of sound in the corresponding channel. c This enables time alignment of all audio channels during interactive behavior, and the timing result of microphone touch in the microphone array is obtained based on the time index of the first sound detection of all channels; the timing result is the touch trajectory. In step S2, the obtained timing result is the interaction category, which includes single-point touch with the first microphone, single-point touch with the second microphone, single-point touch with the third microphone, single-point touch with the fourth microphone, swipe left, swipe right, and invalid trigger.

2. The sound detection method according to claim 1, characterized in that: In step S1, the wake-up time window length is 0.1s and the wake-up sliding step size is 0.1s.

3. The sound detection method according to claim 1, characterized in that: In step S2, after obtaining the timing results, the corresponding interactive instructions are obtained based on the timing results.

4. A sound detection method according to claim 1, characterized in that: In step S2, the root mean square value of each sub-segment of each audio channel segment is calculated according to equation (2). （2） In equation (2), RMS1 is the root mean square value of a single channel audio segment; K is the number of sampling points contained in each sub-segment; and k is the kth sampling point. This is the kth sample value of a mono channel.

5. A sound detection method according to claim 2, characterized in that: In step S2, the technical conditions for time alignment are shown in equation (3). （3） Equation (3) indicates that at the same time, at least one of the first sub-segments of the four channels contains a valid signal, when t c When =None, t c Consider it as +∞. Indicates the channel number.

6. The sound detection method according to claim 1, characterized in that: In step S2, the interaction time window length is 0.2s, the interaction sliding step length is 0.02s, and an audio segment is divided into num sub-segments of equal length, where num=10.

Images