Control method, voice device, control device, electronic device and storage medium

Abstract

The present invention belongs to the technical field of smart home appliances. The present invention provides a control method, a voice device, a control device, an electronic device and a storage medium. The control method is applied to a voice device, including: receiving an audio signal to be processed and extracting a first audio feature; inputting the first audio feature into a neural network model to obtain output data, wherein the neural network model is trained based on a second audio feature of a preset audio signal, and the ratio between the second voice feature of the voice signal in the preset audio signal and the second audio feature of the preset audio signal is the labeled value of the second audio feature; determining the first voice feature and the discrimination value of the voice signal in the audio signal to be processed based on the output data and the first audio feature; uploading the discrimination value, and executing the processing corresponding to the audio signal to be processed if a returned control instruction is received. Through the above-mentioned neural network, it is possible to avoid the non-stationary noise signal from accurately determining the target voice device.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart home appliances, and specifically relates to a control method, a voice device, a control device, an electronic device and a storage medium. Background Art

[0002] At present, voice devices can perform corresponding operations by collecting the user's voice signal. However, when waking up the voice device, if multiple voice devices are set with the same wake-up word, when the user outputs the voice signal, multiple devices may be woken up at the same time.

[0003] In the prior art, since the closer the sound is to the sound source during propagation, the higher the energy of the collected signal is, the target voice device can be determined by comparing the energy of voice signals received by multiple voice devices.

[0004] However, when voice devices collect voice signals, they also collect many non-stationary noise signals, and traditional noise reduction algorithms are difficult to remove these non-stationary noise signals, which makes the energy calculation inaccurate and the final determination of the target voice device has errors. Summary of the invention

[0005] In order to solve the above-mentioned problem in the prior art, that is, to solve the problem of inaccurate determination of the target voice device in the prior art, the present invention provides a control method, a voice device, a control device, an electronic device and a storage medium.

[0006] According to a first aspect of an embodiment of the present application, the present invention provides a control method, applied to a voice device, the method comprising:

[0007] Receiving an audio signal to be processed, and extracting a first audio feature of the audio signal to be processed, wherein the audio signal to be processed is composed of a mixture of a speech signal and a non-stationary noise signal;

[0008] Inputting the first audio feature into a neural network model to obtain output data, wherein the neural network model is trained based on a second audio feature of a preset audio signal, and a ratio between a second speech feature of a speech signal in the preset audio signal and a second audio feature of the preset audio signal is a labeled value of the second audio feature;

[0009] Determining a first speech feature of a speech signal in the audio signal to be processed, and determining a discrimination value corresponding to the first speech feature, based on the output data and the first audio feature of the audio signal to be processed;

[0010] The judgment value is uploaded to the control device, and if a control instruction returned by the control device is received, the processing corresponding to the audio signal to be processed is executed.

[0011] In a preferred technical solution of the above control method, extracting the first audio feature of the audio signal to be processed includes:

[0012] Performing short-time Fourier transform processing on the audio signal to be processed to obtain a first frequency domain signal;

[0013] Determining a first audio feature of the audio signal to be processed according to the first frequency domain signal and a preset Mel frequency conversion formula;

[0014] Determining the first speech feature of the speech signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed includes: determining a second frequency domain signal according to the output data and the preset Mel frequency conversion formula;

[0015] A first speech feature of a speech signal in the audio signal to be processed is determined according to the second frequency domain signal and the first frequency domain signal.

[0016] In the preferred technical solution of the above control method, after determining the first voice feature of the voice signal in the audio signal to be processed based on the output data and the first audio feature of the audio signal to be processed, the method further includes: performing noise reduction processing on the first voice feature according to a preset voice noise reduction algorithm to obtain a noise-reduced signal;

[0017] The determining the discriminant value corresponding to the first speech feature includes:

[0018] Based on the noise-reduced signal and the calibration coefficient of the speech device, a spectrum energy value corresponding to the first speech feature is determined, and the spectrum energy value is used as the discrimination value.

[0019] In the preferred technical solution of the above-mentioned control method, the method also includes: determining a compensation signal based on the output data and a preset noise signal, wherein the noise signal satisfies a Gaussian distribution; performing noise reduction processing on the first voice feature based on a preset speech noise reduction algorithm, including: performing noise reduction processing on a signal obtained by superimposing the compensation signal and the first voice feature based on a preset speech noise reduction algorithm to obtain the noise-reduced signal.

[0020] In the preferred technical solution of the above-mentioned control method, the neural network model includes: a convolutional layer, a recurrent neural network layer, a fully connected layer and an activation layer; inputting the first audio feature into the neural network model to obtain output data includes: inputting the first audio feature into the convolutional layer, the recurrent neural network layer, the fully connected layer and the activation layer connected in sequence, and using the parameters output by the activation layer as the output data.

[0021] In a preferred technical solution of the above control method, before receiving the audio signal to be processed, the method further includes:

[0022] Acquire at least two non-stationary noise signals, and perform superposition processing on them according to a preset ratio to obtain a noise signal after superposition processing; the preset ratio may be determined by the existence duration of the non-stationary noise signal within a preset time period;

[0023] The processed noise signal and the preset voice signal are superimposed to obtain a preset audio signal.

[0024] According to a second aspect of the embodiments of the present application, the present invention provides a control method, which is applied to a control device, and the method includes:

[0025] Receive discrimination values ​​uploaded by multiple voice devices, and determine a target voice device among the multiple voice devices according to a preset discrimination criterion, wherein the discrimination value is that the voice device extracts a first audio feature of the audio signal to be processed received by the voice device, inputs the first audio feature into a neural network model, and obtains output data; and determines a first voice feature of a voice signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed, and determines a discrimination value according to the first voice feature;

[0026] A control instruction is sent to a target voice device, wherein the control instruction is used to instruct the target voice device to respond to a process corresponding to the received audio signal to be processed.

[0027] In the preferred technical solution of the above control method, the discrimination value is a spectrum energy value; determining the target voice device from multiple voice devices based on a preset discrimination criterion includes: selecting a device with the highest spectrum energy value from multiple voice devices as the target voice device.

[0028] According to a third aspect of the embodiments of the present application, the present invention provides a voice device, the voice device comprising:

[0029] A first receiving unit is used to receive an audio signal to be processed and extract a first audio feature of the audio signal to be processed, wherein the audio signal to be processed is composed of a mixture of a speech signal and a non-stationary noise signal;

[0030] A first determining unit is configured to input the first audio feature into a neural network model to obtain output data, wherein the neural network model is trained based on a second audio feature of a preset audio signal, and a ratio between the second speech feature of the speech signal in the preset audio signal and the second audio feature of the preset audio signal is a labeled value of the second audio feature; and determine the first speech feature of the speech signal in the audio signal to be processed based on the output data and the first audio feature of the audio signal to be processed, and determine a discriminant value corresponding to the first speech feature;

[0031] The first sending unit is used to upload the discrimination value to the control device, and execute the processing corresponding to the audio signal to be processed if a control instruction returned by the control device is received.

[0032] In the preferred technical solution of the above-mentioned voice device, the first receiving unit includes:

[0033] A first transformation module, configured to perform short-time Fourier transform processing on the audio signal to be processed to obtain a first frequency domain signal;

[0034] A second transformation module, used to determine a first audio feature of the audio signal to be processed according to the first frequency domain signal and a preset Mel frequency conversion formula;

[0035] The first determining unit includes:

[0036] A first determination module, configured to determine a second frequency domain signal according to the output data and the preset Mel frequency conversion formula;

[0037] The second determination module is used to determine a first speech feature of a speech signal in the audio signal to be processed according to the second frequency domain signal and the first frequency domain signal.

[0038] In the preferred technical solution of the above-mentioned voice device, the voice device further includes: a noise reduction unit, which is used to perform noise reduction processing on the first voice feature according to a preset voice noise reduction algorithm to obtain a noise-reduced signal after the first determination unit determines the first voice feature of the voice signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed;

[0039] The first determination unit is specifically configured to determine, based on the noise-reduced signal and a calibration coefficient of the speech device, a spectrum energy value corresponding to the first speech feature, and use the spectrum energy value as the discrimination value.

[0040] In the preferred technical solution of the above-mentioned voice device, the voice device further includes:

[0041] A second determination unit is used to determine a compensation signal according to the output data and a preset noise signal before the noise reduction unit performs noise reduction processing on the first speech feature according to a preset speech noise reduction algorithm and uses the processed signal as the first speech feature of the speech signal, wherein the noise signal satisfies a Gaussian distribution;

[0042] The noise reduction unit is specifically used to: perform noise reduction processing on the signal obtained by superimposing the compensation signal and the first voice feature according to a preset voice noise reduction algorithm to obtain the noise-reduced signal.

[0043] In the preferred technical solution of the above-mentioned voice device, the neural network model includes: a convolutional layer, a recurrent neural network layer, a fully connected layer and an activation layer; the first determination unit is specifically used to input the first audio feature into the convolutional layer, the recurrent neural network layer, the fully connected layer and the activation layer connected in sequence, and use the parameters output by the activation layer as the output data.

[0044] In the preferred technical solution of the above-mentioned voice device, the voice device further includes:

[0045] an acquisition unit, configured to acquire at least two non-stationary noise signals before the first receiving unit receives the audio signal to be processed, and perform superposition processing on the two signals in accordance with a preset ratio to obtain a noise signal after superposition processing; the preset ratio may be determined by the existence duration of the non-stationary noise signal in a preset time period;

[0046] The processing unit is used to perform superposition processing on the processed noise signal and the preset voice signal to obtain a preset audio signal.

[0047] According to a fourth aspect of the embodiments of the present application, the present invention provides a control device, the control device comprising:

[0048] A second receiving unit is used to receive the discrimination values ​​uploaded by multiple voice devices, and determine the target voice device among the multiple voice devices according to a preset discrimination criterion, wherein the discrimination value is that the voice device extracts a first audio feature of the audio signal to be processed received by the voice device, and inputs the first audio feature into the neural network model to obtain output data; and according to the output data and the first audio feature of the audio signal to be processed, determines the first voice feature of the voice signal in the audio signal to be processed, and determines the discrimination value according to the first voice feature;

[0049] The second sending unit is used to send a control instruction to the target voice device, where the control instruction is used to instruct the target voice device to respond to the processing corresponding to the received audio signal to be processed.

[0050] In a preferred technical solution of the above control device, the discrimination value is a spectrum energy value; and the second receiving unit is specifically configured to select a device with a highest spectrum energy value from a plurality of voice devices as a target voice device.

[0051] According to a fifth aspect of the embodiments of the present application, the present invention provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0052] The memory stores computer-executable instructions;

[0053] The processor executes the computer-executable instructions stored in the memory to implement the control method as described in any one of the first aspects of the embodiments of the present application.

[0054] According to a sixth aspect of the embodiments of the present application, the present invention provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0055] The memory stores computer-executable instructions;

[0056] The processor executes the computer-executable instructions stored in the memory to implement the control method as described in any one of the second aspects of the embodiments of the present application.

[0057] According to the seventh aspect of the embodiments of the present application, the present invention provides a computer-readable storage medium, which stores computer execution instructions. When the computer execution instructions are executed by a processor, they are used to implement the control method as described in any one of the first aspects of the embodiments of the present application.

[0058] According to the eighth aspect of the embodiments of the present application, the present invention provides a computer-readable storage medium, in which computer execution instructions are stored. When the computer execution instructions are executed by a processor, they are used to implement the control method as described in any one of the second aspects of the embodiments of the present application.

[0059] According to a ninth aspect of the embodiments of the present application, the present invention provides a computer program product, including a computer program, which is executed by a processor as the control method described in any one of the first aspect of the embodiments of the present application.

[0060] According to the tenth aspect of the embodiments of the present application, the present invention provides a computer program product, including a computer program, which is executed by a processor as the control method described in any one of the first aspect of the embodiments of the present application.

[0061] It can be understood by those skilled in the art that the control method, speech device, control device, electronic device and storage medium provided by the present invention are applied to speech devices, including: receiving an audio signal to be processed, and extracting a first audio feature of the audio signal to be processed, wherein the audio signal to be processed is composed of a mixture of a speech signal and a non-stationary noise signal; inputting the first audio feature into a neural network model to obtain output data, wherein the neural network model is trained based on the second audio feature of a preset audio signal, and the ratio between the second speech feature of the speech signal in the preset audio signal and the second audio feature of the preset audio signal is the labeled value of the second audio feature; determining the first speech feature of the speech signal in the audio signal to be processed based on the output data and the first audio feature of the audio signal to be processed, and determining the discrimination value corresponding to the first speech feature; uploading the discrimination value to the control device, and executing the processing corresponding to the audio signal to be processed if a control instruction returned by the control device is received. Through the above-mentioned neural network, the influence of the non-stationary noise signal on the accuracy of determining the target speech device can be avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The preferred embodiments of the control method, voice device, control device, electronic device and storage medium of the present invention are described below with reference to the accompanying drawings, which are as follows:

[0063] Figure 1 A schematic diagram of an application scenario provided for this application;

[0064] Figure 2 A flow chart of a control method provided in an embodiment of the present application;

[0065] Figure 3 A schematic diagram of a process for extracting audio signal features provided in an embodiment of the present application;

[0066] Figure 4 A flowchart of another control method applied to a voice device provided in an embodiment of the present application;

[0067] Figure 5 A flowchart of another control method applied to a voice device provided in an embodiment of the present application;

[0068] Figure 6 A flowchart of a training method for a neural network model provided in an embodiment of the present application;

[0069] Figure 7 A flow chart of a control method applied to a control device provided in an embodiment of the present application;

[0070] Figure 8 An interactive schematic diagram of a control method provided in an embodiment of the present application;

[0071] Fig. 9 A schematic diagram of the structure of a voice device provided in an embodiment of the present application;

[0072] Fig.10 A structural diagram of another voice device provided in an embodiment of the present application;

[0073] Fig.11 A schematic diagram of the structure of a control device provided in an embodiment of the present application;

[0074] Fig.12 A schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0075] First, those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention. Those skilled in the art can make adjustments to them as needed to adapt to specific application scenarios.

[0076] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "connected" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0077] First, the terms involved in the present invention are explained:

[0078] 1) Smart home appliances refer to home appliances that are formed by introducing microprocessors, sensor technology, and network communication technology into home appliances. They have the characteristics of intelligent control, intelligent perception, and intelligent application. The operation process of smart home appliances often relies on the application and processing of modern technologies such as the Internet of Things, the Internet, and electronic chips. For example, smart home appliances can be connected to electronic devices to enable users to remotely control and manage smart home appliances.

[0079] 2) Terminal device refers to an electronic device with wireless connection function. The terminal device can communicate with the above-mentioned smart home appliances by connecting to the Internet, or directly communicate with the above-mentioned smart home appliances through Bluetooth, WiFi, etc. In some embodiments, the terminal device is, for example, a mobile device, a computer, or a vehicle-mounted device built into a hover car, or any combination thereof. Mobile devices may include, for example, mobile phones, smart home devices, wearable devices, smart mobile devices, virtual reality devices, etc., or any combination thereof, wherein wearable devices include, for example: smart watches, smart bracelets, pedometers, etc.

[0080] 3) "Multiple" means two or more, and other quantifiers are similar. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the related objects are in an "or" relationship.

[0081] 4) "Correspondence" can refer to an association relationship or a binding relationship. The correspondence between A and B means that there is an association relationship or a binding relationship between A and B.

[0082] The application scenarios of the embodiments of the present application are explained below:

[0083] At present, with the development of technology, more and more smart home appliances have voice control functions. In the same scenario, when multiple smart home appliances are equipped with voice wake-up modules, when the user says the wake-up word, multiple devices will respond to the user's voice control at the same time, but in fact the user only wants to interact with one device.

[0084] An example, to solve the above problem, such as Figure 1 A schematic diagram of an application scenario provided by the present application. In the figure, multiple smart home appliances (the smart home appliances here can be washing machines, range hoods, dryers, sweeping robots, smart speakers, etc.) can be connected to the same control device (the control device here can be a remote server or cloud). When each smart home appliance collects a voice signal output by a user, the energy of the voice signal can be extracted, and then the energy of the analyzed voice signal is uploaded to the control device. The control device then compares and judges based on the energy uploaded by the multiple voice devices received, and the control device determines the target home appliance (i.e., the smart home appliance 3 in the figure), and sends a determination result to the determined smart home appliance so that the smart home appliance responds to the processing corresponding to the user's voice signal.

[0085] However, when smart home appliances collect user voice signals, they usually carry some environmental noise, which can be divided into stationary noise signals and non-stationary noise signals. The single-channel noise reduction algorithm in the related art can only remove the stationary noise signal in the collected audio signal, but cannot remove the non-stationary noise signal. As a result, the noise signal in the collected signal will affect the smart home appliance's judgment of the signal energy, which will make the control device's final determination of the target home appliance inaccurate.

[0086] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0087] Figure 2 A flow chart of a control method provided in an embodiment of the present application is shown, and the method is applied to a voice device, such as Figure 2 As shown, the control method provided in this embodiment includes the following steps:

[0088] Step S201: receiving an audio signal to be processed and extracting a first audio feature of the audio signal to be processed, wherein the audio signal to be processed is composed of a mixture of a speech signal and a non-stationary noise signal.

[0089] For example, the voice device in this embodiment can be a device controlled by voice, such as a smart washing machine, a smart speaker, a sweeping robot, etc. In addition, the control device can be connected to the Internet to communicate with the above-mentioned voice device, or can be directly communicated with the above-mentioned voice device through Bluetooth, WiFi, etc.

[0090] Specifically, the speech device in this embodiment can be used to collect the audio signal to be processed in the environment, wherein the audio signal to be processed is composed of a mixture of the speech signal output by the user and the non-stationary noise signal. When the speech device collects the above-mentioned audio signal to be processed, it will perform speech feature extraction on the audio signal to be processed.

[0091] In a possible implementation, the time of the audio signal to be processed in the time domain and the signal amplitude corresponding to each time may be used as the first audio feature of the audio signal to be processed.

[0092] In another possible implementation, since the user's voice signal and the noise signal are relatively easy to distinguish in the frequency domain, the frequency of the audio signal to be processed in the frequency domain and the amplitude corresponding to the frequency can be used as the first audio feature of the audio signal to be processed.

[0093] Step S202: input the first audio feature into the neural network model to obtain output data, wherein the neural network model is trained based on the second audio feature of the preset audio signal, and the ratio between the second speech feature of the speech signal in the preset audio signal and the second audio feature of the preset audio signal is the labeled value of the second audio feature.

[0094] Exemplarily, in order to remove the non-stationary noise signal contained in the collected audio signal to be processed, the first audio feature obtained in step S201 can be input into a pre-trained neural network model.

[0095] Among them, when training, the neural network model is obtained by presetting the second audio feature of the audio signal, and taking the ratio of the second voice feature of the voice signal in the preset audio signal to the second audio feature of the preset audio signal as the label value of the second audio feature to train the model. Specifically, in an example, the ratio of the second voice feature to the second audio feature can be used as the objective function in the training process, and the loss function is obtained by comparing the data output each time during the training process with the objective function. When the loss value calculated by the loss function is less than the preset value, the training ends and a trained neural network model is obtained.

[0096] Step S203: determining a first speech feature of a speech signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed, and determining a discrimination value corresponding to the first speech feature.

[0097] Exemplarily, after obtaining the output data of the neural network model, that is, the ratio between the first speech feature and the first audio feature, the first speech feature of the speech signal in the audio signal to be processed can be directly calculated based on the ratio and the characteristic value of the input audio signal to be processed.

[0098] After the first speech feature is determined, a feature value of the speech signal may be calculated based on the first speech feature.

[0099] In one example, when the first audio feature of the input is the time domain amplitude feature of the audio signal to be processed, the first voice feature obtained is the amplitude feature of the voice signal in the time domain, and then the discrimination value corresponding to the voice signal, such as the energy value, can be calculated based on the amplitude feature, or the distance between the user and the voice device can be estimated.

[0100] In another example, when the first audio feature of the input is the amplitude feature of the audio signal to be processed in the frequency domain, the first voice feature obtained is the amplitude feature of the voice signal in the frequency domain, and then the discrimination value corresponding to the voice signal, such as the energy value, can be calculated based on the amplitude feature, or the distance between the user and the voice device can be estimated.

[0101] Step S204: Upload the judgment value to the control device. If a control instruction returned by the control device is received, execute the processing corresponding to the audio signal to be processed.

[0102] Exemplarily, after the voice device calculates the discrimination value, it uploads the discrimination value to the corresponding control device. After the control device receives the discrimination values ​​uploaded by multiple voice devices, it selects a target voice device from the multiple voice devices according to a preset discrimination rule. If the voice device receives the control instruction sent by the control device, it indicates that the voice device is selected as the target voice device, and the voice device performs the processing corresponding to the audio signal to be processed.

[0103] In one example, if the voice device does not receive the control instruction sent by the control device within a preset time, it indicates that the voice device is not the target voice device selected by the user, and the voice device does not perform any operation.

[0104] For example, when a user needs to use a smart washing machine to wash clothes, he sends a voice command "turn on" to the smart washing machine. At this time, the smart speakers and sweeping robots in the same space receive the voice command at the same time. In addition, when each device collects the user's voice information, it also collects a lot of environmental noise. Therefore, each device extracts the features of the collected audio signal to be processed and inputs it into the pre-stored neural network model. After model processing, the voice features of the voice signal collected by the device and the corresponding discrimination value of the feature are determined. Each device uploads its own discrimination value. After judgment by the control device, the voice device that receives the control signal sent by the control device executes the user's voice command "turn on", and the other voice devices do not perform the boot process.

[0105] In this embodiment, after the voice device receives the audio signal to be processed, the signal can be feature extracted. At this time, the time domain features or frequency domain features of the audio signal to be processed can be selected and input into a pre-trained model. Compared with extracting the time domain features, since the difference between the features of the clean voice signal and the noise signal in the time domain is small, and it is easy to distinguish within the frequency domain, the features in the time domain of the audio signal to be processed can be selected to input into the neural network model. Or in order to avoid the processing process when extracting the frequency domain features, the time domain features can also be directly selected. Afterwards, based on the model output, the first voice feature is extracted from the audio signal to be processed, that is, the audio signal to be processed is processed to remove non-stationary noise, avoiding the influence of the non-stationary noise signal on the determination process of the target voice device, so that the first voice feature corresponding to the discrimination value of the voice signal determined by the subsequent voice device is more accurate, thereby improving the accuracy of the judgment of the target voice device.

[0106] In one embodiment, Figure 3 A flowchart of an audio signal feature extraction method provided in an embodiment of the present application is shown in FIG. Figure 2 When extracting the first audio feature of the audio signal to be processed in step S201 in the embodiment, the following steps may be adopted:

[0107] Step S301: Perform short-time Fourier transform processing on the audio signal to be processed to obtain a first frequency domain signal.

[0108] Step S302: Determine a first audio feature of the audio signal to be processed according to the first frequency domain signal and a preset Mel frequency conversion formula.

[0109] Exemplarily, when extracting features from an audio signal to be processed, first, the audio signal to be processed can be subjected to a short-time Fourier transform, and the audio signal to be processed can be converted from the time domain to the frequency domain to obtain a first frequency domain signal. Since the human ear itself has a series of critical frequency bands, the sound in the same critical frequency band is prone to a masking effect, that is, the sound signal in the critical frequency band is easily masked by another signal with large energy and close frequency, resulting in a person being unable to perceive the sound signal. Moreover, the human ear is more sensitive to low-frequency signals than high-frequency signals. Therefore, the above-mentioned frequency domain signal can be converted into a critical frequency band, that is, the amplitude of the frequency domain signal is converted into a Bark domain signal through a preset Mel frequency (i.e., Mel frequency) conversion formula, and the amplitude of the Bark domain signal is used as the first audio feature of the audio signal to be processed, and then through the above-mentioned conversion process, the low-frequency signal is widened and the high-frequency signal is compressed.

[0110] In this embodiment, the signal obtained after the audio signal to be processed is processed by short-time Fourier transform and Mel frequency conversion as the first audio feature of the audio signal to be processed. Compared with directly using the first frequency domain signal after short-time Fourier transform as the first audio feature of the audio signal to be processed, since the signal will become a signal with symmetrical amplitude after short-time Fourier transform, when performing Mel frequency mapping, only half of the amplitude in the original signal needs to be mapped, and the amount of data after Mel mapping is reduced compared to the first frequency domain signal, thereby reducing the amount of data of the model input feature and reducing the complexity of model training.

[0111] Specifically, when using Figure 3 When using the feature extraction method shown, the data output by the neural network model is also Bark domain data, so it is necessary to convert the output data in the Bark domain to obtain output data in the frequency domain, that is, the signal dimension will be compressed after step S302, so when obtaining the output data of the neural network model in the Bark domain, it is also necessary to convert the output data to the frequency domain.

[0112] That is, when executing Figure 2 When determining the first voice feature of the voice signal in the audio signal to be processed based on the output data and the first audio feature of the audio signal to be processed, the following steps can be used to implement:

[0113] The first step is to determine the second frequency domain signal according to the output data and the preset Mel frequency conversion formula;

[0114] Exemplarily, in this step, when the input feature of the neural network model is a Bark domain feature that has undergone Mel frequency conversion, at this time, the data output by the neural network model needs to be converted from the Bark domain to the frequency domain according to a preset Mel frequency conversion formula to obtain a second frequency domain signal corresponding to the output data.

[0115] The second step is to determine the first speech feature of the speech signal in the audio signal to be processed according to the second frequency domain signal and the first frequency domain signal.

[0116] Exemplarily, the second frequency domain signal obtained in the first step is the ratio between the first speech feature of the speech signal in the audio signal to be processed and the first audio feature (here, the first frequency domain signal in step S301). The first speech feature of the speech signal can be obtained through this ratio and the first frequency domain signal.

[0117] In some embodiments, since the audio signal to be processed collected by the speech device includes not only speech signals and non-stationary noise signals, but also some stationary noise signals, after the non-stationary noise signals in the audio signal to be processed are removed by the neural network model, the signal can be further subjected to noise reduction processing to remove the stationary noise signals in the audio signal to be processed. Figure 4 A flow chart of another control method applied to a voice device provided in an embodiment of the present application is as follows: Figure 4 As shown, the method comprises the following steps:

[0118] Step S401: receiving an audio signal to be processed, and extracting a first audio feature of the audio signal to be processed.

[0119] Step S402: input the first audio feature into the neural network model to obtain output data, wherein the neural network model is trained by using the second audio feature of the preset audio signal as the neural network model input and using the ratio of the second speech feature of the speech signal in the preset audio signal to the second audio feature of the preset audio signal as the target output of the neural network model.

[0120] Step S403: Determine a first speech feature of a speech signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed.

[0121] For example, the specific principles of steps S401 to S403 in this embodiment are similar to those in Figure 2 The principles of steps S201-S203 in are similar and will not be repeated here.

[0122] Step S404: performing noise reduction processing on the first speech feature according to a preset speech noise reduction algorithm to obtain a noise-reduced signal.

[0123] Exemplarily, after obtaining the first speech feature through the output data and the first audio feature of the audio signal to be processed, the first speech feature can continue to be subjected to noise reduction processing (for example, Wiener filtering, OMLSA (optimally-modified log-spectral amplitude, optimally improved logarithmic spectral amplitude estimation) algorithm, etc.) to remove the stationary noise signal in the first speech feature, thereby obtaining a signal after noise reduction processing.

[0124] Step S405: Determine the spectrum energy value corresponding to the first speech feature based on the noise-reduced signal and the calibration coefficient of the speech device, and use the spectrum energy value as a discrimination value.

[0125] For example, in actual application scenarios, since the audio signal acquisition devices carried by various voice devices have different sensitivities to sound signals, even if the distance between the user and the audio signal acquisition devices is the same and the voice signals input by the user are the same, the signals collected by the audio signal acquisition devices are different, and thus the calculated spectrum energy values ​​are also different, which affects the judgment of the target voice device. Therefore, when the voice device calculates the spectrum energy corresponding to the voice feature, it is also necessary to calibrate the calculated energy value according to the calibration coefficient of the voice device and send the calibrated energy value to the control device.

[0126] Specifically, when setting the calibration coefficient for each voice device, in one implementation, the audio signal acquisition devices in multiple voice devices can be placed in the same environment (that is, the distance between each acquisition device and the sound source must be kept consistent, and the audio emitted by the sound source to each acquisition device is the same). Then, the spectral energy value calculated from the audio signal collected by one of the acquisition devices is used as a benchmark, and the ratio of the spectral energy of the remaining acquisition devices to the spectral energy is used as a preset calibration coefficient.

[0127] In this step, after obtaining the denoised signal, the spectral energy value of the denoised signal is determined, and the spectral energy value of the denoised signal and the spectral energy value determined by the calibration coefficient of the speech device are used as the spectral energy value corresponding to the first speech feature, and then the above spectral energy value is used as the discrimination value.

[0128] Step S406: Upload the judgment value to the control device. If a control instruction returned by the control device is received, execute the processing corresponding to the audio signal to be processed.

[0129] For example, the principle of step S406 is the same as Figure 2 The principle in step S203 is similar and will not be repeated here.

[0130] In this embodiment, after obtaining the first speech feature of the speech signal through the neural network model, the first speech feature is also subjected to traditional stationary noise denoising processing, thereby avoiding affecting the accuracy of the target device selection when the collected audio signal to be processed is mixed with a stationary noise signal. In addition, by setting a calibration coefficient for each speech device, it is avoided that when the sensitivity of the audio acquisition device in different speech devices is different, the final judgment value uploaded to the control device is inaccurate, thereby affecting the selection of the target speech device.

[0131] In some embodiments, Figure 5 A flow chart of another control method for a voice device provided in an embodiment of the present application. As shown in the figure, the method includes the following steps:

[0132] Step S501: receiving an audio signal to be processed, and extracting a first audio feature of the audio signal to be processed.

[0133] Step S502: input the first audio feature into the neural network model to obtain output data, wherein the neural network model is trained by using the second audio feature of the preset audio signal as the neural network model input and using the ratio of the second speech feature of the speech signal in the preset audio signal to the second audio feature of the preset audio signal as the target output of the neural network model.

[0134] Step S503: Determine a first speech feature of a speech signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed.

[0135] For example, the specific principles of steps S501 to S503 in this embodiment are the same as those in Figure 2 The principles of steps S201-S203 in are similar and will not be repeated here.

[0136] Step S504: determining a compensation signal according to the output data and a preset noise signal, wherein the noise signal satisfies a Gaussian distribution.

[0137] Exemplarily, in this embodiment, when the non-stationary noise signal in the processed audio signal is subjected to noise reduction processing by the neural network model, the stationary component in the non-stationary noise is also suppressed, while the stationary component in the remaining processed audio signal interval without non-stationary noise still exists, which will cause the noise spectrum to be discontinuous, thereby affecting the accuracy of the noise reduction processing for the stationary noise. Therefore, before the noise reduction processing of the stationary noise is performed, the compensation signal is determined to be superimposed on the first speech feature obtained by the neural network model through step S504.

[0138] Specifically, the compensation signal can be obtained by using the output data of the neural network model and the preset noise signal. In practical applications, the compensation signal is a signal obtained by doing a dot product between the output data of the model and the preset noise signal. And the preset noise signal is a stationary signal, that is, it needs to satisfy the Gaussian distribution.

[0139] Step S505: performing noise reduction processing on the signal obtained by superimposing the compensation signal and the first speech feature according to a preset speech noise reduction algorithm to obtain a noise-reduced signal.

[0140] Exemplarily, after the first speech feature is converted into a continuous signal by superimposing the first speech feature with the compensation signal, the continuous signal may be subjected to noise reduction processing to remove stationary noise in the first speech feature to obtain a noise-reduced signal.

[0141] Step S506: Determine the spectrum energy value corresponding to the first speech feature based on the noise-reduced signal and the calibration coefficient of the speech device, and use the spectrum energy value as a discrimination value.

[0142] Step S507: Upload the judgment value to the control device. If a control instruction returned by the control device is received, execute the processing corresponding to the audio signal to be processed.

[0143] For example, the principles of step S506 and step S507 are the same as Figure 4 The principles of step S405 and step S406 are similar and will not be repeated here.

[0144] In this embodiment, in order to avoid the discontinuity of the noise spectrum in the first voice feature obtained by the neural network model, which may lead to inaccurate noise reduction results when the stationary noise is subsequently processed, it is necessary to perform superposition processing on the first voice feature, that is, to generate a compensation signal based on the preset noise signal and the output data of the neural network model, and perform superposition processing on the first voice feature to compensate for the stationary components in the non-stationary noise signal removed by the neural network model. Afterwards, the superimposed signal is subjected to noise reduction processing, and the discrimination value is determined based on the signal after noise reduction processing to improve the accuracy of the target voice device selection.

[0145] In some embodiments, the neural network model includes: a convolutional layer, a recurrent neural network layer, a fully connected layer and an activation layer; in step S202, it specifically includes: inputting the first audio feature into the convolutional layer, the recurrent neural network layer, the fully connected layer and the activation layer connected in sequence, and using the parameters output by the activation layer as output data.

[0146] Among them, the convolution layer is used to compress the data of the input first audio feature. For example, multiple frames of data can be compressed into one frame of data, and the compressed data includes the features of multiple frames of data, thereby reducing the amount of data processing in model training or application. The recurrent neural network layer can adopt a traditional recurrent neural network (RNN) or LSTM or GRU, without specific restrictions here. In addition, the fully connected layer in the model can be used to reduce the dimension of the output features of the recurrent neural network layer to the same dimension as the features of the convolution layer input. The activation layer after the fully connected layer can be used to map the features output by the fully connected layer to an interval with a smaller value range, such as an interval of (0-1), by selecting an activation function, so as to reduce the complexity of the model calculation.

[0147] For example: the neural network model structure can be:

[0148] The first layer is a one-dimensional convolutional layer:

[0149] Input parameter dimension in_channels = 64, output parameter dimension out_channels = 64, convolution kernel size kernel_size = 4;

[0150] The second layer is the LSTM layer:

[0151] Input parameter dimension Input_size = 64, hidden node number hidden_size = 80;

[0152] Finally there are fully connected layers and activation layers.

[0153] Input parameter dimension Input_size = 80, hidden node number hidden_size = 64

[0154] Finally, the neural network model outputs: a 64*1 vector.

[0155] Figure 6 A flowchart of a training method for a neural network model provided in an embodiment of the present application is shown below. Figure 6 As shown, the method comprises the following steps:

[0156] Step S601, obtaining at least two non-stationary noise signals, and superimposing them according to a preset ratio to obtain a noise signal after superimposition processing; the preset ratio can be determined by the existence duration of the non-stationary noise signal in a preset time period.

[0157] Exemplarily, after acquiring at least two non-stationary noise signals, the acquired non-stationary noise signals may be mixed according to a preset ratio. The preset ratio may be determined according to the duration of existence of the non-stationary noise signal in a preset time period. For example, if the working duration of the range hood is longer than that of the sweeping robot in a day, the preset ratio value of the range hood may be greater than the preset ratio value of the sweeping robot.

[0158] Step S602: Superimpose the processed noise signal and the preset voice signal to obtain a preset audio signal.

[0159] Exemplarily, after step S601, the processed noise signal obtained in step S601 and a preset voice signal are superimposed and processed, where the preset voice signal may be a power-on voice command issued by a user, thereby obtaining a preset audio signal.

[0160] In one example, when performing superposition processing of a noise signal and a speech signal, a preset signal-to-noise ratio range needs to be met, that is, the processed noise signal and the preset speech signal are superimposed according to the preset signal-to-noise ratio range to obtain a preset audio signal.

[0161] In this example, when the noise signal and the voice signal are superimposed, that is, when executing step S602, specifically, the processed noise signal and the preset voice signal need to be superimposed according to a preset signal-to-noise ratio range to avoid interference with the voice signal when the noise signal accounts for a large proportion. The larger the signal-to-noise ratio, the smaller the proportion of the noise signal in the superimposed signal. For example, the signal-to-noise ratio range can be set to [-10, 15] dB.

[0162] In this embodiment, when performing model training, the obtained multiple non-stationary noise signals can be superimposed, and when superimposing, the preset ratio can be determined according to the duration of the non-stationary noise signal in the preset time period, so that the non-stationary noise signal is superimposed according to the preset ratio, so that the superimposed noise signal is closer to the noise signal in the real environment. And when superimposing the noise signal and the voice signal, the superimposition processing can be performed according to the preset signal-to-noise ratio range, so as to avoid affecting the recognition of the voice signal when the noise signal is large.

[0163] Figure 7 A flow chart of a control method for controlling a device provided in an embodiment of the present application, the method comprising the following steps:

[0164] Step S701, receiving discrimination values ​​uploaded by multiple voice devices, and determining a target voice device among the multiple voice devices according to a preset discrimination criterion, wherein the discrimination value is that the voice device extracts a first audio feature of the audio signal to be processed received by the voice device, and inputs the first audio feature into a neural network model to obtain output data; and based on the output data and the first audio feature of the audio signal to be processed, determines the first voice feature of the voice signal in the audio signal to be processed, and determines the discrimination value based on the first voice feature.

[0165] Step S702: Send a control instruction to the target voice device, where the control instruction is used to instruct the target voice device to respond to the processing corresponding to the received audio signal to be processed.

[0166] Exemplarily, the control method in this embodiment is applied to a control device, wherein the control device can communicate with the voice device in the above embodiment, and the control device can be a controller or a cloud server, etc. The control device can be used to receive the discrimination values ​​uploaded by multiple voice devices, and select a target voice device from multiple voice devices based on the received discrimination values ​​and pre-set discrimination criteria. In addition, when the target voice device is determined, a control instruction can be sent to the target voice device so that the target voice device responds to the audio signal to be processed received by it.

[0167] In one example, when the discrimination value is the spectrum energy value, the control device can select the device with the highest spectrum energy value from multiple voice devices as the target voice device, that is, the preset discrimination criterion is to select the device with the highest spectrum energy value. That is, it is believed that the higher the spectrum energy, the closer the distance between the voice device and the user.

[0168] In this embodiment, the control device judges the target voice device by receiving the discrimination value corresponding to the signal processed by the neural network model. Since the neural network model can suppress the non-stationary signal in the signal output by it, the non-stationary signal will not be affected when calculating the discrimination value, thereby improving the accuracy of the target voice device judgment.

[0169] Figure 8 The interactive schematic diagram of a control method provided in an embodiment of the present application. As shown in the figure, the scene includes a voice device 1, a voice device 2 and a control device. The method includes the following steps:

[0170] Step S801: The voice device receives an audio signal to be processed and extracts a first audio feature of the audio signal to be processed, where the audio signal to be processed is composed of a mixture of a voice signal and a non-stationary noise signal.

[0171] Step S802: The voice device inputs the first audio feature into the neural network model to obtain output data.

[0172] Step S803: The speech device determines a first speech feature of a speech signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed, and determines a discrimination value corresponding to the first speech feature.

[0173] Step S804: The voice device uploads the discrimination value to the control device.

[0174] Step S805: The control device receives the discrimination values ​​uploaded by the multiple voice devices, and determines the target voice device among the multiple voice devices according to a preset discrimination criterion.

[0175] Step S806: Send a control instruction to the target voice device. The target voice device determined in the figure is voice device 1.

[0176] Step S807: If a control instruction returned by the control device is received, the processing corresponding to the audio signal to be processed is performed.

[0177] For example, steps S801-S807 in this embodiment can be referred to Figure 2 and Figure 7 The principles of each step in the process will not be repeated here.

[0178] Fig. 9 A schematic diagram of the structure of a voice device provided in an embodiment of the present application, such as Fig. 9 As shown, the voice device provided in this embodiment includes:

[0179] The first receiving unit 51 is used to receive an audio signal to be processed and extract a first audio feature of the audio signal to be processed, where the audio signal to be processed is composed of a mixture of a speech signal and a non-stationary noise signal;

[0180] A first determining unit 52 is configured to input the first audio feature into a neural network model to obtain output data, wherein the neural network model is trained based on the second audio feature of a preset audio signal, and the ratio between the second voice feature of the voice signal in the preset audio signal and the second audio feature of the preset audio signal is the labeled value of the second audio feature; determine the first voice feature of the voice signal in the audio signal to be processed based on the output data and the first audio feature of the audio signal to be processed, and determine the discrimination value corresponding to the first voice feature;

[0181] The first sending unit 53 is used to upload the discrimination value to the control device, and execute the processing corresponding to the audio signal to be processed if a control instruction returned by the control device is received.

[0182] Fig.10 A structural diagram of another voice device provided in an embodiment of the present application is shown in FIG. Fig.10 As shown, in Fig. 9 In the preferred technical scheme of the voice device based on the device shown,

[0183] The first receiving unit 51 includes:

[0184] A first transform module 511, configured to perform short-time Fourier transform processing on the audio signal to be processed to obtain a first frequency domain signal;

[0185] A second transformation module 512, configured to determine a first audio feature of the audio signal to be processed according to the first frequency domain signal and a preset Mel frequency conversion formula;

[0186] The first determining unit 52 includes:

[0187] A first determination module 521, configured to determine a second frequency domain signal according to the output data and a preset Mel frequency conversion formula;

[0188] The second determination module 522 is configured to determine a first speech feature of a speech signal in the audio signal to be processed according to the second frequency domain signal and the first frequency domain signal.

[0189] In the preferred technical solution of the above-mentioned speech device, the speech device further includes: a noise reduction unit 54, which is used to perform noise reduction processing on the first speech feature according to a preset speech noise reduction algorithm to obtain a noise-reduced signal after the first determination unit 52 determines the first speech feature of the speech signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed,

[0190] The first determination unit is specifically configured to determine, based on the noise-reduced signal and a calibration coefficient of the speech device, a spectrum energy value corresponding to the first speech feature, and use the spectrum energy value as a discrimination value.

[0191] In the preferred technical solution of the above-mentioned voice device, the voice device further includes:

[0192] The second determination unit 55 is used to determine the compensation signal according to the output data and the preset noise signal before the noise reduction unit 54 performs noise reduction processing on the first speech feature according to the preset speech noise reduction algorithm and uses the processed signal as the first speech feature of the speech signal, wherein the noise signal satisfies the Gaussian distribution;

[0193] The noise reduction unit is specifically used to: perform noise reduction processing on a signal obtained by superimposing the compensation signal and the first voice feature according to a preset voice noise reduction algorithm to obtain a noise-reduced signal.

[0194] In the preferred technical solution of the above-mentioned voice device, the neural network model includes: a convolutional layer, a recurrent neural network layer, a fully connected layer and an activation layer; the first determination unit 52 is specifically used to input the first audio feature into the convolutional layer, the recurrent neural network layer, the fully connected layer and the activation layer connected in sequence, and use the parameters output by the activation layer as output data.

[0195] In the preferred technical solution of the above-mentioned voice device, the voice device further includes:

[0196] The acquisition unit 56 is used to acquire at least two non-stationary noise signals before the first receiving unit 51 receives the audio signal to be processed, and perform superposition processing on them according to a preset ratio to obtain a noise signal after superposition processing; the preset ratio can be determined by the existence time of the non-stationary noise signal in a preset time period;

[0197] The processing unit 57 is used to perform superposition processing on the processed noise signal and the preset voice signal to obtain a preset audio signal.

[0198] Fig.11 A schematic diagram of a control device provided in an embodiment of the present application is shown in FIG. Fig.11 As shown, the control device provided in this embodiment includes:

[0199] The second receiving unit 61 is used to receive the discrimination values ​​uploaded by multiple voice devices, and determine the target voice device among the multiple voice devices according to a preset discrimination criterion, wherein the discrimination value is that the voice device extracts the first audio feature of the audio signal to be processed received by the voice device, and inputs the first audio feature into the neural network model to obtain output data; and determines the first voice feature of the voice signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed, and determines the discrimination value according to the first voice feature;

[0200] The second sending unit 62 is used to send a control instruction to the target voice device, where the control instruction is used to instruct the target voice device to respond to the processing corresponding to the received audio signal to be processed.

[0201] In the preferred technical solution of the above control device, the discrimination value is the spectrum energy value; the second receiving unit 61 is specifically used to select the device with the highest spectrum energy value from multiple voice devices as the target voice device.

[0202] For related instructions, please refer to Figure 2-Figure 6 The relevant descriptions and effects corresponding to the steps in the corresponding embodiments can be understood, and no further elaboration is made here.

[0203] The control device provided in this embodiment can perform the following operations: Figure 2-Figure 6The technical solution of the method embodiment shown has similar implementation principles and technical effects, which will not be repeated here.

[0204] Fig.12 A schematic diagram of an electronic device provided in an embodiment of the present application, such as Fig.12 As shown, the electronic device provided in this embodiment includes: a memory 701, a processor 702 and a computer program.

[0205] The computer program is stored in the memory 701 and is configured to be executed by the processor 702 to implement the present invention. Figure 2-Figure 10 The control method provided in any one of the corresponding embodiments.

[0206] The memory 701 and the processor 702 are connected via a bus 703 .

[0207] For related instructions, please refer to Figure 2-Figure 6 The relevant descriptions and effects corresponding to the steps in the corresponding embodiments can be understood, and no further elaboration is made here.

[0208] One embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the present invention. Figure 2-Figure 6 The control method provided in any one of the corresponding embodiments.

[0209] One embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the present invention. Figure 7 The corresponding embodiment provides a control method.

[0210] Among them, the computer-readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

[0211] The present invention provides a computer program product, comprising a computer program, which is executed by a processor to implement the present invention. Figure 2-Figure 6 The control method provided in any one of the corresponding embodiments.

[0212] The present invention provides a computer program product, comprising a computer program, which is executed by a processor to implement the present invention. Figure 7 The control method provided in the corresponding embodiment.

[0213] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0214] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the applications disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The description and examples are to be considered exemplary only, and the true scope and spirit of the present invention is indicated by the following claims.

[0215] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

[0216] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0217] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the applications disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The description and examples are to be considered exemplary only, and the true scope and spirit of the present invention is indicated by the following claims.

Claims

1. A control method, characterized in that: Applied to a voice device, the method comprises: Receive an audio signal to be processed, and extract a first audio feature of the audio signal to be processed, wherein the audio signal to be processed is composed of a mixture of a speech signal and a non-stationary noise signal; wherein the first audio feature is the time of the audio signal to be processed in the time domain and the signal amplitude corresponding to each time, or the first audio feature is the frequency of the audio signal to be processed in the frequency domain and the amplitude corresponding to the frequency; Inputting the first audio feature into a neural network model to obtain output data, wherein the neural network model is trained based on a second audio feature of a preset audio signal, and a ratio between a second speech feature of a speech signal in the preset audio signal and a second audio feature of the preset audio signal is a labeled value of the second audio feature; Determining a first speech feature of a speech signal in the audio signal to be processed, and determining a discrimination value corresponding to the first speech feature, based on the output data and the first audio feature of the audio signal to be processed; Uploading the discrimination value to the control device, and executing the processing corresponding to the audio signal to be processed if a control instruction returned by the control device is received; The method further comprises: Determine a compensation signal according to the output data and a preset noise signal, wherein the noise signal satisfies a Gaussian distribution; the compensation signal is a signal obtained by performing a dot product of the output data of the model and the preset noise signal; Before receiving the audio signal to be processed, the method further includes: Acquire at least two non-stationary noise signals, and perform superposition processing on them according to a preset ratio to obtain a noise signal after superposition processing; the preset ratio is determined by the existence duration of the non-stationary noise signal within a preset time period; The processed noise signal and the preset voice signal are superimposed according to a preset signal-to-noise ratio range to obtain a preset audio signal.

2. The method according to claim 1, characterized in that The extracting the first audio feature of the audio signal to be processed includes: Performing short-time Fourier transform processing on the audio signal to be processed to obtain a first frequency domain signal; Determining a first audio feature of the audio signal to be processed according to the first frequency domain signal and a preset Mel frequency conversion formula; The determining, based on the output data and the first audio feature of the audio signal to be processed, the first speech feature of the speech signal in the audio signal to be processed comprises: Determining a second frequency domain signal according to the output data and the preset Mel frequency conversion formula; A first speech feature of a speech signal in the audio signal to be processed is determined according to the second frequency domain signal and the first frequency domain signal.

3. The method according to claim 1, characterized in that After determining the first speech feature of the speech signal in the audio signal to be processed according to the output data and the first audio feature of the audio signal to be processed, the method further includes: Performing noise reduction processing on the first speech feature according to a preset speech noise reduction algorithm to obtain a noise-reduced signal; The determining the discriminant value corresponding to the first speech feature includes: Based on the noise-reduced signal and the calibration coefficient of the speech device, a spectrum energy value corresponding to the first speech feature is determined, and the spectrum energy value is used as the discrimination value.

4. The method according to claim 3, characterized in that The performing noise reduction processing on the first speech feature according to a preset speech noise reduction algorithm includes: The signal obtained by superimposing the compensation signal and the first speech feature is subjected to noise reduction processing according to a preset speech noise reduction algorithm to obtain the noise-reduced signal.

5. The method according to claim 1, characterized in that The neural network model includes: a convolutional layer, a recurrent neural network layer, a fully connected layer, and an activation layer; the step of inputting the first audio feature into the neural network model to obtain output data includes: The first audio feature is input into the convolutional layer, the recurrent neural network layer, the fully connected layer and the activation layer which are connected in sequence, and the parameters output by the activation layer are used as the output data.

6. A control method, characterized in that: Applied to a control device, the method comprises: Receive discrimination values ​​uploaded by multiple voice devices, and determine a target voice device among the multiple voice devices according to a preset discrimination criterion, wherein the discrimination value is a first audio feature of an audio signal to be processed extracted by the voice device, and the first audio feature is input into a neural network model to obtain output data; wherein the neural network model is trained based on a second audio feature of a preset audio signal, and a ratio between the second voice feature of a voice signal in the preset audio signal and the second audio feature of the preset audio signal is a labeled value of the second audio feature; Determine a compensation signal based on the output data and a preset noise signal, wherein the noise signal satisfies a Gaussian distribution; the compensation signal is a signal obtained by performing a dot product between the output data of the model and the preset noise signal; and determine a first speech feature of a speech signal in the audio signal to be processed based on the output data, the preset noise signal and a first audio feature of the audio signal to be processed, and determine a discrimination value based on the first speech feature; wherein the first audio feature is the time of the audio signal to be processed in the time domain and the signal amplitude corresponding to each time, or the first audio feature is the frequency of the audio signal to be processed in the frequency domain and the amplitude corresponding to the frequency; Sending a control instruction to a target voice device, wherein the control instruction is used to instruct the target voice device to perform processing corresponding to the received audio signal to be processed; Among them, the preset audio signal is obtained by superimposing a noise signal obtained by superimposing multiple non-stationary noise signals according to a preset ratio and a preset voice signal according to a preset signal-to-noise ratio range; the preset ratio is determined by the existence duration of the non-stationary noise signal within a preset time period.

7. A voice device, characterized in that: include: A first receiving unit is used to receive an audio signal to be processed and extract a first audio feature of the audio signal to be processed, wherein the audio signal to be processed is composed of a mixture of a speech signal and a non-stationary noise signal; wherein the first audio feature is the time of the audio signal to be processed in the time domain and the signal amplitude corresponding to each time, or the first audio feature is the frequency of the audio signal to be processed in the frequency domain and the amplitude corresponding to the frequency; A first determining unit is configured to input the first audio feature into a neural network model to obtain output data, wherein the neural network model is trained based on a second audio feature of a preset audio signal, and a ratio between the second speech feature of the speech signal in the preset audio signal and the second audio feature of the preset audio signal is a labeled value of the second audio feature; and determine the first speech feature of the speech signal in the audio signal to be processed based on the output data and the first audio feature of the audio signal to be processed, and determine a discriminant value corresponding to the first speech feature; A first sending unit, configured to upload the discrimination value to a control device, and execute processing corresponding to the audio signal to be processed if a control instruction returned by the control device is received; A second determination unit is used to determine a compensation signal according to the output data and a preset noise signal, wherein the noise signal satisfies a Gaussian distribution; and the compensation signal is a signal obtained by performing a dot product between the output data of the model and the preset noise signal; An acquisition unit, used for acquiring at least two non-stationary noise signals, and performing superposition processing on them according to a preset ratio to obtain a noise signal after superposition processing; the preset ratio is determined by the existence duration of the non-stationary noise signal within a preset time period; The processing unit is used to perform superposition processing on the processed noise signal and the preset voice signal according to a preset signal-to-noise ratio range to obtain a preset audio signal.

8. A control device, characterized in that: include: A second receiving unit is used to receive discrimination values ​​uploaded by multiple voice devices, and determine a target voice device from the multiple voice devices according to a preset discrimination criterion, wherein the discrimination value is a first audio feature of the audio signal to be processed extracted by the voice device, and the first audio feature is input into a neural network model to obtain output data; wherein the neural network model is trained based on a second audio feature of a preset audio signal, and a ratio between the second voice feature of the voice signal in the preset audio signal and the second audio feature of the preset audio signal is a labeled value of the second audio feature; Determine a compensation signal based on the output data and a preset noise signal, wherein the noise signal satisfies a Gaussian distribution; the compensation signal is a signal obtained by dot product of the output data of the model and the preset noise signal; and determine a first voice feature of a voice signal in the audio signal to be processed based on the output data, the preset noise signal and the first audio feature of the audio signal to be processed, and determine a discrimination value based on the first voice feature; wherein the first audio feature is the time of the audio signal to be processed in the time domain and the signal amplitude corresponding to each time, or the first audio feature is the frequency of the audio signal to be processed in the frequency domain and the amplitude corresponding to the frequency; wherein the preset audio signal is obtained by superimposing a noise signal obtained by superimposing multiple non-stationary noise signals according to a preset ratio and a preset voice signal according to a preset signal-to-noise ratio range; the preset ratio is determined by the existence duration of the non-stationary noise signal in a preset time period; The second sending unit is used to send a control instruction to the target voice device, where the control instruction is used to instruct the target voice device to perform processing corresponding to the received audio signal to be processed.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 5.

Images