DAS-based audio time difference prediction method and system, electronic device and storage medium

Abstract

The application relates to the technical field of optical fiber sensing, in particular to an audio time difference prediction method and system based on DAS, an electronic device and a storage medium, which comprises the following steps: obtaining audio digital signals of two audio channels according to sound source signals collected by DAS; respectively extracting features of the audio digital signals of the two audio channels to obtain audio feature vectors corresponding to the audio digital signals of the audio channels; respectively identifying features of the two audio feature vectors to obtain sound source events corresponding to the audio feature vectors; if the two sound source events are the same and belong to a target event, then the two audio feature vectors are spliced to obtain a spliced feature vector; and predicting an audio time domain offset according to the spliced feature vector. Compared with the prior art, the application can exclude the interference of non-target events and reduce the calculation consumption of useless events or signals with low signal-to-noise ratio which are difficult to use for calculation by comparing the sound source events identified by the two audio channels with the target event.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical fiber sensing, and more particularly to an audio time difference prediction method and system based on DAS, an electronic device and a storage medium. BACKGROUND

[0002] Optical cable sound source positioning mainly uses the time difference of sound waves or vibrations generated by sound source events reaching the optical cable to realize the positioning of sound source events. Therefore, audio time difference calculation of the optical cable is an important technology in optical cable sound source positioning. However, existing audio time difference calculation, such as the binaural audio time difference algorithm, needs to rely on accurate audio data. However, the signal collected by DAS (Distributed Optical Fiber Acoustic Sensing) is affected by the continuous sensing characteristics, and the signal-to-noise ratio is lower than that of the audio signal collected by conventional audio acquisition equipment, which leads to the fact that the existing audio time difference calculation cannot adaptively adjust to some interfering sound source events when applied to DAS, thereby reducing the calculation accuracy and also wasting computing resources. SUMMARY

[0003] The present application provides an audio time difference prediction method and system based on DAS, an electronic device and a storage medium, which are used to improve the accuracy of audio time difference prediction of DAS and reduce the consumption of computing resources.

[0004] According to a first aspect of the present application, an audio time difference prediction method based on DAS is provided, the method comprising:

[0005] obtaining audio digital signals of two audio channels according to the sound source signals collected by DAS; the audio channels are divided by the optical cable according to a preset channel length;

[0006] respectively extracting features of the audio digital signals of the two audio channels to obtain audio feature vectors corresponding to the audio digital signals of each audio channel;

[0007] respectively identifying features of the two audio feature vectors to obtain sound source events corresponding to each audio feature vector;

[0008] if the two sound source events are the same and belong to a target event, then splicing the two audio feature vectors to obtain a spliced feature vector;

[0009] predicting an audio time domain offset according to the spliced feature vector.

[0010] The application simulates two audio channels as "binaural", collects audio digital signals containing sound source signals of the two audio channels based on DAS, and realizes positioning of sound source events. In one aspect, the audio digital signals of the two audio channels are subjected to feature recognition to identify sound source events. The feature recognition can more accurately capture sound source events of the optical cable. When the sound source events of the two audio channels of the optical cable are the same, the consistency of "binaural" can be better determined, the confusion of event types can be avoided, and the accuracy of positioning of sound source events can be improved. In another aspect, when the two sound source events are the same and belong to target events, the audio time domain offset is predicted based on the splicing feature vector, the interference of non-target events can be excluded, the calculation consumption of useless events or events with low signal-to-noise ratio that are difficult to calculate can be reduced, and the audio time difference prediction of the sound source events of the optical cable can be more accurately realized based on the features of the sound source events in the two audio channels.

[0011] Optionally, the feature extraction of the audio digital signals of the two audio channels respectively to obtain the audio feature vectors corresponding to the audio digital signals of each audio channel comprises:

[0012] The audio digital signals of the two audio channels are input into a twin neural network for feature extraction to obtain the audio feature vectors corresponding to the audio digital signals of each audio channel. One subnetwork of the twin neural network corresponds to the feature extraction of the audio digital signals of one audio channel.

[0013] The feature processing of the audio digital signals of the two audio channels by the twin neural network can well combine the features of the audio digital signals in the two audio channels for recognition, thereby improving the accuracy of the subsequent prediction of the audio time domain offset.

[0014] Optionally, the subnetwork of the twin neural network comprises an LSTM network, an RNN network, a regular network and a nonlinear processing network.

[0015] One subnetwork of the twin neural network corresponds to the acquisition of the audio feature vectors of the audio digital signals of one audio channel, comprising:

[0016] The audio digital signals are subjected to feature extraction by the LSTM network to obtain first audio time sequence features;

[0017] The first audio time sequence features are subjected to feature processing by the RNN network to obtain second audio time sequence features;

[0018] The second audio time sequence features are subjected to regular processing by the regular network to obtain regularized features;

[0019] The regularized features are subjected to nonlinear conversion by the nonlinear processing network to obtain the audio feature vectors.

[0020] The LSTM network effectively extracts the time sequence features in the audio digital signal to obtain the first audio time sequence features, and the RNN network further processes the first audio time sequence features, further captures the time sequence dependency relationship through the memory information captured by the LSTM network, so that the finally obtained audio feature vector can effectively contain the features of the time sequence in the audio digital signal, thereby improving the accuracy of predicting the audio time domain offset.

[0021] Optionally, the audio time domain offset is predicted according to the spliced feature vector, comprising:

[0022] The convolution network is used to perform convolution processing on the spliced feature vector to obtain a convolution vector;

[0023] The full connection output network is used to process the convolution vector to output the audio time domain offset.

[0024] The convolution network is used to effectively extract the audio features to effectively predict the spliced feature vector and improve the accuracy of predicting the audio time domain offset.

[0025] Optionally, the sound source event includes a main sound source event and a sub-sound source event.

[0026] The main sound source event is a sound source event type of the sound source signal.

[0027] The sub-sound source event is a vibration state of the main sound source event, and the sub-sound source event at least includes a vibration starting state, a vibration state, and a vibration ending state.

[0028] By further dividing the sound source event into a main sound source event and a sub-sound source event, it can be further determined whether to predict the audio time domain offset through the vibration state of the sound source event, and the audio digital signal corresponding to the vibration state with higher signal-to-noise ratio can be used to predict the audio time domain offset, thereby further improving the accuracy of predicting the audio time domain offset.

[0029] Optionally, the target event includes a main target event and a sub-target event; and the sub-target event includes a vibration starting state and a vibration ending state.

[0030] The two sound source events are the same and belong to the target event, including:

[0031] The main sound source event and the sub-sound source event in the two sound source events are the same, and the main sound source event belongs to the main target event, and the sub-sound source event is the vibration starting state or the vibration ending state.

[0032] The audio digital signals of the onset state and the end state in the vibration state of the sound source event have higher signal-to-noise ratios, and can be better used for prediction of audio time domain offsets, and filter out vibration states with interference.

[0033] Optionally, the audio digital signals of the two audio channels are obtained according to the sound source signals collected by the DAS, and the method comprises:

[0034] The audio channel in which the position point of the optical cable of the sound source signal collected by the DAS is located is taken as a first audio channel;

[0035] An audio channel adjacent to the first audio channel is taken as a second audio channel;

[0036] The audio digital signals of the first audio channel and the second audio channel are respectively obtained.

[0037] The use of the two adjacent audio channels can improve the correlation of the features in the collected audio digital signals, and further improve the accuracy of audio time difference prediction.

[0038] According to a second aspect of the present application, a DAS-based audio time difference prediction system is provided, and the system comprises:

[0039] A signal acquisition module is configured to obtain audio digital signals of two audio channels according to sound source signals collected by the DAS; the audio channels are divided by an optical cable according to a preset channel length;

[0040] A feature acquisition module is configured to respectively extract features from the audio digital signals of the two audio channels, and obtain audio feature vectors corresponding to the audio digital signals of the audio channels;

[0041] An event identification module is configured to respectively identify the features of the two audio feature vectors, and obtain sound source events corresponding to the audio feature vectors;

[0042] A feature splicing module is configured to splice the two audio feature vectors to obtain a spliced feature vector if the two sound source events are the same and belong to a target event;

[0043] An offset prediction module is configured to predict an audio time domain offset according to the spliced feature vector.

[0044] According to a third aspect of the present application, an electronic device is provided, and the electronic device comprises:

[0045] A memory is configured to store one or more computer programs;

[0046] A processor is configured to implement the DAS-based audio time difference prediction method of the first aspect when the one or more computer programs are executed by the processor.

[0047] According to a fourth aspect of the present application, a computer readable storage medium is provided, which stores computer instructions for causing a processor to implement the DAS-based audio time difference prediction method of the first aspect when executed.

[0048] Based on any one of the above aspects, the DAS-based audio time difference prediction method, system, electronic device and computer readable storage medium provided by the embodiments of the present application realize positioning of a sound source event by simulating "binaural" for two audio channels, and collecting audio digital signals containing a sound source signal of the two audio channels based on DAS. In one aspect, the audio digital signals of the two audio channels are subjected to feature recognition to identify a sound source event. The sound source event of the optical cable can be more accurately captured by the feature recognition method. In this case, when the sound source events of the two audio channels of the optical cable are the same, the consistency of "binaural" can be better determined, the confusion of event types can be avoided, and the accuracy of positioning of the sound source event can be improved. In another aspect, when the two sound source events are the same and belong to a target event, the spliced feature vector is used to predict an audio time domain offset. Non-target events can be excluded, the calculation consumption of useless events or low signal-to-noise ratio events that are difficult to calculate can be reduced, and the audio time difference prediction of the sound source event of the optical cable can be more accurately realized by using the features of the sound source event in the two audio channels. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0050] Figure 1 The step flowchart of the audio time difference prediction method provided by the present embodiment.

[0051] Figure 2 The model structure diagram of the twin neural network provided by the present embodiment.

[0052] Figure 3 The step flowchart of the audio feature vector acquisition provided by the present embodiment.

[0053] Figure 4 The step flowchart of the audio digital signal acquisition provided by the present embodiment.

[0054] Figure 5 The system structure diagram of the audio time difference prediction system provided by the present embodiment.

[0055] Figure 6 A device structure diagram of an electronic device is provided for the embodiment.

[0056] The figure caption is: twin neural network 100, LSTM network 110, RNN network 120, regular network 130, nonlinear processing network 140, signal acquisition module 11, feature acquisition module 12, event recognition module 13, feature splicing module 14, offset prediction module 15, memory 21, processor 22, communication module 23, input / output interface 24, bus 25. DETAILED DESCRIPTION

[0057] The drawings of the present application are only used for illustrative description, and cannot be understood as a limitation of the present application. In order to better illustrate the following embodiments, some components of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; it is understandable for those skilled in the art that some well-known structures and their descriptions in the drawings may be omitted.

[0058] In order to make the person skilled in the art better understand the scheme of the present application, the technical scheme in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0059] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0060] The optical cable sound source positioning mainly utilizes the time difference of the sound waves or vibrations generated by the sound source event to reach the optical cable to realize the positioning of the sound source event. Therefore, the audio time difference calculation of the optical cable is an important technology in the optical cable sound source positioning. The existing audio time difference calculation methods include the calculation method based on generalized cross-correlation, the frequency domain sub-band decomposition and phase analysis method, and the binaural coherence function calculation method. However, these existing audio time difference calculation methods have different calculation methods, but generally need to rely on accurate audio data for corresponding calculation. In the optical cable, the DAS is used to obtain the audio signal of the optical cable. The optical cable is continuously affected by the sound source. Influenced by this continuous sensing characteristic, the signal-to-noise ratio of the audio signal obtained by the DAS is lower than that of the audio signal collected by the conventional audio acquisition device (such as a microphone), which leads to that when the existing audio time difference calculation is applied to the DAS, it cannot adaptively adjust to some interfering sound source events, thereby reducing the calculation accuracy and also causing waste of calculation resources.

[0061] The embodiment provides a technical solution that can solve the above problems. The specific embodiments of the application will be described in detail below with reference to the accompanying drawings.

[0062] As shown in Figure 1 The embodiment provides an audio time difference prediction method based on DAS, which can include the following steps:

[0063] S1: obtaining audio digital signals of two audio channels according to the sound source signals collected by the DAS;

[0064] In the embodiment, when a sound source event occurs around the optical cable, the optical cable will receive the sound wave signals generated by the sound source event. When the optical cable receives the sound wave signals, it will generate corresponding scattered light. The DAS device receives the scattered light and analyzes the scattered light to obtain an audio analog signal. The audio analog signal is obtained through pulse code modulation to obtain the corresponding audio digital signal as the sound source signal.

[0065] It can be understood that the optical cable position point can be the sampling point that first receives the sound wave signal on the optical cable. Since the sound wave signal of the sound source event is propagated to the surrounding, a certain range of the optical cable will be affected by the sound wave signal, and corresponding audio digital signals will be generated in the range. The audio time difference prediction of the sound source event needs at least two groups of audio digital signals.

[0066] To this end, the embodiment divides the optical cable into a plurality of audio channels in advance, and when the sound wave signal of the sound source event reaches the optical cable, the prediction of the audio time difference is quickly realized by acquiring the audio digital signal generated by the audio channel affected by the sound wave signal. Wherein, the audio channel is divided by the optical cable according to the preset channel length. For example, if the length of the optical cable is 10km and the preset channel length is 100m, the optical cable can be divided into 100 audio channels.

[0067] In an alternative embodiment, as shown in Figure 4 The step S1 can include the following steps:

[0068] S11: Taking the audio channel in which the optical cable position point of the sound source signal collected by the DAS is located as the first audio channel;

[0069] S12: Taking one audio channel adjacent to the first audio channel as the second audio channel;

[0070] S13: Acquiring the audio digital signal of the first audio channel and the second audio channel respectively.

[0071] Taking the audio channel of the optical cable position point as the first audio channel and the audio channel adjacent thereto as the second audio channel, the two audio channels obtained have higher correlation in data and can better simulate "binaural" to achieve accurate positioning. In addition, when the first audio channel is acquired, the second audio channel is directly determined according to the first audio channel, which can reduce the arrival judgment of the optical cable position point and reduce the consumption of calculation.

[0072] S2: Extracting features from the audio digital signals of the two audio channels respectively to obtain the audio feature vectors corresponding to the audio digital signals of each audio channel;

[0073] In the embodiment, the feature extraction of the audio digital signal in the step S2 can be:

[0074] Inputting the audio digital signals of the two audio channels into a twin neural network to extract features and obtain the audio feature vectors corresponding to the audio digital signals of each audio channel; wherein, one subnetwork of the twin neural network corresponds to the feature extraction of the audio digital signal of one audio channel.

[0075] In an embodiment, as shown in Figure 2As shown, the twin neural network 100 can include an LSTM (Long Short-Term Memory Network) network 110, an RNN (Recurrent Neural Network) network 120, a regularization network 130, and a nonlinear processing network 140.

[0076] In the embodiment, as shown, Figure 3 The step S2 can include the following steps:

[0077] S21: performing feature extraction on the audio digital signal by the LSTM network to obtain first audio time sequence features;

[0078] S22: performing feature processing on the first audio time sequence features by the RNN network to obtain second audio time sequence features;

[0079] S23: performing regularization processing on the second audio time sequence features by the regularization network to obtain regularization features;

[0080] S24: performing nonlinear conversion on the regularization features by the nonlinear processing network to obtain the audio feature vector.

[0081] The LSTM network 110 can effectively capture features in long time sequence data through the memory unit and gating mechanism therein. Therefore, in the embodiment, the LSTM network 110 is used to extract time sequence features of the continuously input audio digital signal, capture features of the continuous sound wave signal emitted by the sound source event, and obtain the first audio time sequence features.

[0082] The RNN network 120 can pass feature information at a previous moment to a next moment by memorizing historical information in sequence data, so that the continuous previous sequence features can affect the subsequent sequence features. Therefore, in the embodiment, the RNN network 120 is used to perform feature processing on the first audio time sequence features extracted by the LSTM network 110 to obtain the second audio time sequence features. In this process, information in the first audio time sequence features at a previous time sequence is passed to a next time sequence, which can better capture the correlation and difference between different time sequences, and further effectively extract features of the sound source event from the audio digital signal of the DAS with low signal-to-noise ratio, which is conducive to identifying the sound source event type of the sound source event.

[0083] The regularization network 130 can include a Drop-out network unit, which is used to perform regularization processing on the second audio time sequence features obtained by the RNN network 120 to obtain regularization features, avoid overfitting in the learning process, and improve the generalization ability of the network.

[0084] The nonlinear processing network 140 can include two nonlinear processing layers, each of which can include a Linear network unit and a ReLU activation function for nonlinear conversion of the input features, wherein the Linear network unit performs linear conversion of the input features, and the ReLU activation function introduces nonlinearity to the linearly transformed features, which can improve the processing capability of the network for complex data.

[0085] S3: respectively performing feature recognition on the two audio feature vectors to obtain sound source events corresponding to the audio feature vectors;

[0086] In this embodiment, the audio feature vectors in step S3 can be processed by a Linear network unit respectively to obtain the sound source events. Since the sound source events have been very comprehensively learned in the subnetwork of the twin neural network 100, the obtained audio feature vectors can be directly linearly converted by the Linear network unit to recognize the sound source events.

[0087] S4: if the two sound source events are the same and belong to a target event, then splicing the two audio feature vectors to obtain a spliced feature vector;

[0088] In this embodiment, the sound source events can include main sound source events and sub-sound source events; wherein the main sound source events are sound source event types of the sound source signal, such as construction site construction and vehicle passing, etc., and the sub-sound source events are vibration states of the main sound source events, which can at least include a vibration state, a vibration state, and a vibration end state.

[0089] On the contrary, the target event can include a main target event and a sub-target event, the main target event corresponds to the main sound source event, and contains several sound source event types that need to predict the audio time difference, and the sub-target event includes a vibration state and a vibration end state.

[0090] It can be understood that some sound source event types have less impact on the optical cable, and such sound source event types do not need to predict the audio time difference for positioning. Therefore, by setting the main target event of the target event, the embodiment can filter out the sound source events that need to predict the audio time difference, reduce the calculation amount that needs to be predicted, and reduce the computational burden.

[0091] And the sound source event includes the whole process of the event activity from the pre-vibration to the end of the vibration; therefore, the whole process of the event activity can be divided into multiple vibration states. Among them, the signal-to-noise ratio of the audio digital signal obtained in the pre-vibration state and the end of vibration state is high, and the signal-to-noise ratio of the audio digital signal obtained in other vibration states such as the bottom noise vibration and the saturated vibration state is low, and using these vibration states for prediction will result in a large error of the audio time difference obtained.

[0092] Therefore, the embodiment further divides the sound source event into a main sound source event and a sub-sound source event, and filters through the main target event and the sub-target event of the target event. Based on the sound source event type, the sound source event that needs to be predicted for the audio time difference can be filtered, and based on the vibration state of the sound source event type, the audio feature vector of the corresponding vibration interval of the sound source event can be filtered, so that the embodiment can pay more attention to the pre-vibration state and the end of vibration state of the sound source event, and the features extracted from the audio digital signal with high signal-to-noise ratio in the pre-vibration state and the end of vibration state are used to predict the audio time difference, reduce the interference of various noises, and effectively improve the prediction accuracy.

[0093] S5: predicting an audio time domain offset according to the spliced feature vector.

[0094] In the embodiment, the spliced feature vector contains the features of the audio digital signals of the adjacent two audio channels, so that the audio time domain offset can be predicted through the spliced feature vector.

[0095] In an implementation, the step S5 can include the following steps:

[0096] performing convolution processing on the spliced feature vector through a convolution network to obtain a convolution vector;

[0097] performing processing on the convolution vector through a fully connected output network to output the audio time domain offset.

[0098] The convolution network can be set as a convolution neural network unit, and the fully connected output network can be set as a Linear network unit.

[0099] In the embodiment, the audio time domain offset is obtained by using the deep neural network prediction method. Compared with the traditional audio time difference calculation method, the embodiment can better capture the time domain correlation of the audio digital signals of the adjacent audio channels in the DAS, and thus can more accurately predict the audio time domain offset.

[0100] The embodiment also provides an audio time difference prediction system based on the DAS, as shown in Figure 5 The system can include:

[0101] The signal acquisition module 11 is configured to acquire audio digital signals of two audio channels according to the sound source signals collected by the DAS.

[0102] In this embodiment, the signal acquisition module 11 is configured to perform step S1 as shown in the following table. Figure 1 The specific content of the signal acquisition module 11 can refer to the specific description of step S1.

[0103] The feature acquisition module 12 is configured to perform feature extraction on the audio digital signals of the two audio channels respectively, and acquire audio feature vectors corresponding to the audio digital signals of the two audio channels respectively.

[0104] In this embodiment, the feature acquisition module 12 is configured to perform step S2 as shown in the following table. Figure 1 The specific content of the feature acquisition module 12 can refer to the specific description of step S2.

[0105] The event recognition module 13 is configured to perform feature recognition on the two audio feature vectors respectively, and acquire sound source events corresponding to the two audio feature vectors respectively.

[0106] In this embodiment, the event recognition module 13 is configured to perform step S3 as shown in the following table. Figure 1 The specific content of the event recognition module 13 can refer to the specific description of step S3.

[0107] The feature splicing module 14 is configured to splice the two audio feature vectors to obtain a spliced feature vector if the two sound source events are the same and belong to a target event.

[0108] In this embodiment, the feature splicing module 14 is configured to perform step S4 as shown in the following table. Figure 1 The specific content of the feature splicing module 14 can refer to the specific description of step S4.

[0109] The offset prediction module 15 is configured to predict an audio time domain offset according to the spliced feature vector.

[0110] In this embodiment, the offset prediction module 15 is configured to perform step S5 as shown in the following table. Figure 1 The specific content of the offset prediction module 15 can refer to the specific description of step S5.

[0111] The embodiments of the present application also provide an electronic device, the structure of which is shown in the following table. Figure 6 The embodiments of the present application also provide an electronic device, the structure of which is shown in the following table.

[0112] The electronic device includes a memory 21, a processor 22, a communication module 23, and an input / output interface 24, etc. Optionally, the memory 21, the processor 22, the communication module 23, and the input / output interface 24 can be connected and communicated through a bus 25.

[0113] The memory 21 is configured to store one or more computer programs and transmit codes of the computer programs to the processor 22; when the one or more computer programs are executed by the processor 22, the DAS-based audio time difference prediction method in the embodiments of the present application is implemented.

[0114] Optionally, the electronic device can be connected to a network through the communication module 23 to communicate with other devices such as terminals or servers through the network to realize data interaction. The electronic device can be various forms of digital computers, such as desktop computers, servers, workstations, mainframe computers, or other types of computers. The electronic device can also be various forms of mobile terminals, such as smartphones, tablet computers, wearable devices (such as helmets, glasses, watches, etc.), and other similar mobile terminals.

[0115] Optionally, the electronic device can connect the required input / output devices such as keyboards, display devices, etc. through the input / output interface 24. The electronic device itself can have a display device, and other display devices can also be connected through the input / output interface 24. Optionally, storage devices such as hard disks, etc. can also be connected through the input / output interface 24, so that data in the electronic device can be stored in the storage device, or data in the storage device can be read, and data in the storage device can also be stored in the memory 21. It can be understood that the input / output interface 24 can be a wired interface or a wireless interface. According to different actual application scenarios, the devices connected to the input / output interface 24 can be a component of the electronic device, or an external device connected to the electronic device when needed.

[0116] Optionally, the memory 21 can be a volatile memory and / or a non-volatile memory. The volatile memory can be a random access memory, etc. The non-volatile memory can be a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, or a flash memory, etc.

[0117] Optionally, the computer program stored in the processor 22 can be divided into one or more modules stored in the memory 21 and executed by the processor 22 to complete the method provided by the embodiment. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0118] Optionally, the processor 22 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 22 include, but are not limited to, a central processing unit, a graphics processing unit, a digital signal processor, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, and any appropriate controller, microcontroller, processor, etc. The processor 22 executes various methods and processes of the embodiment, exemplarily, a DAS-based audio time difference prediction method of the embodiment.

[0119] Optionally, the bus 25 can include a channel for transmitting information. According to different functions, the bus 25 can be divided into an address bus, a data bus, a control bus, etc.

[0120] In an optional implementation, the embodiment further provides a computer storage medium having a computer program stored thereon, which enables the computer to execute the method of the method embodiment when executed by the computer. Part or all of the computer program can be loaded and / or installed on the memory 21 of the electronic device. When the computer program is executed by the processor 22, one or more steps of the DAS-based audio time difference prediction method of the embodiment can be executed.

[0121] Optionally, the computer readable storage medium can be a random access memory, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, etc.

[0122] Obviously, the above embodiments of the present application are only examples for clearly illustrating the technical solutions of the present application, and are not intended to limit the specific embodiments of the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the claims of the present application shall be included in the protection scope of the claims of the present application.

Claims

1. A DAS-based audio time difference prediction method, characterized by, The method comprises: According to the sound source signal collected by the DAS, the audio digital signals of two audio channels are obtained; the audio channels are divided by the optical cable according to the preset channel length; The audio digital signals of the two audio channels are respectively subjected to feature extraction to obtain the audio feature vectors corresponding to the audio digital signals of each audio channel; The two audio feature vectors are respectively subjected to feature recognition to obtain the sound source events corresponding to each audio feature vector; If the two sound source events are the same and belong to the target event, the two audio feature vectors are spliced to obtain a spliced feature vector; The audio time domain offset is predicted according to the spliced feature vector.

2. The DAS-based audio time difference prediction method of claim 1, wherein, The feature extraction of the audio digital signals of the two audio channels respectively to obtain the audio feature vectors corresponding to the audio digital signals of each audio channel comprises: The audio digital signals of the two audio channels are input into a twin neural network for feature extraction to obtain the audio feature vectors corresponding to the audio digital signals of each audio channel; wherein one subnetwork of the twin neural network corresponds to the audio feature vector of the audio digital signal of one audio channel.

3. The DAS-based audio time difference prediction method of claim 2, wherein, The subnetwork of the twin neural network comprises an LSTM network, an RNN network, a regular network and a nonlinear processing network; One subnetwork of the twin neural network corresponds to the audio feature vector of the audio digital signal of one audio channel, comprising: The audio digital signal is subjected to feature extraction through an LSTM network to obtain a first audio time sequence feature; The first audio time sequence feature is subjected to feature processing through an RNN network to obtain a second audio time sequence feature; The second audio time sequence feature is subjected to regular processing through a regular network to obtain a regularized feature; The regularized feature is subjected to nonlinear conversion through a nonlinear processing network to obtain the audio feature vector.

4. The DAS-based audio time difference prediction method of claim 1, wherein, The prediction of the audio time domain offset according to the spliced feature vector comprises: The spliced feature vector is subjected to convolution processing through a convolution network to obtain a convolution vector; The convolution vector is processed through a fully connected output network to output the audio time domain offset.

5. The DAS-based audio time difference prediction method according to any one of claims 1-4, characterized in that, The sound source event comprises a main sound source event and a sub-sound source event; The main sound source event is the sound source event type of the sound source signal; The sub-sound source event is the vibration state of the main sound source event, and the sub-sound source event at least comprises a vibration starting state, a vibration state and a vibration ending state.

6. The DAS-based audio time difference prediction method of claim 5, wherein, The target event comprises a main target event and a sub-target event; wherein the sub-target event comprises a vibration starting state and a vibration ending state; The two sound source events are the same and belong to the target event, comprising: The main sound source event and the sub-sound source event in the two sound source events are the same, and the main sound source event belongs to the main target event, and the sub-sound source event is the vibration starting state or the vibration ending state.

7. The DAS-based audio time difference prediction method according to any one of claims 1-4, characterized in that, The audio digital signals of two audio channels are obtained according to the sound source signal collected by the DAS, comprising: The audio channel at the optical cable position point of the sound source signal collected by the DAS is taken as a first audio channel; An audio channel adjacent to the first audio channel is taken as a second audio channel; Acquire audio digital signals of the first audio channel and the second audio channel respectively.

8. A DAS-based audio time difference prediction system, characterized by, The system comprises: A signal acquisition module is configured to acquire audio digital signals of two audio channels according to sound source signals collected by the DAS; the audio channels are divided by an optical cable according to a preset channel length; A feature acquisition module is configured to extract features from the audio digital signals of the two audio channels respectively, and acquire audio feature vectors corresponding to the audio digital signals of each audio channel; An event identification module is configured to identify features of the two audio feature vectors respectively, and acquire sound source events corresponding to each audio feature vector; A feature splicing module is configured to splice the two audio feature vectors to obtain a spliced feature vector if the two sound source events are the same and belong to a target event; An offset prediction module is configured to predict an audio time domain offset according to the spliced feature vector.

9. An electronic device, comprising: Comprise: A memory is configured to store one or more computer programs; A processor, when the one or more computer programs are executed by the processor, implements the DAS-based audio time difference prediction method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, and the computer instructions are used to make the processor execute the DAS-based audio time difference prediction method according to any one of claims 1-7.

Images