A voice wake-up method, device, electronic device, medium and program product

Abstract

The present disclosure relates to a voice wake-up method, device, electronic device, medium and program product, the method comprising: when the current state of the voice wake-up function is a dormant state, identifying whether the current ambient sound includes a wake-up voice to obtain a first-level recognition result; when the first-level recognition result indicates that the current ambient sound includes a wake-up voice, sending the current ambient sound to a server; receiving a second-level recognition result returned by the server, in which the server identifies whether the current ambient sound meets a target wake-up condition; when it is determined that the second-level recognition result indicates that the current ambient sound meets the target wake-up condition, switching the current state from a dormant state to a wake-up state. The present disclosure combines the first-level recognition result and the second-level recognition result to determine whether to switch the state of the voice wake-up function to a wake-up state, thereby ensuring a high accuracy rate of voice wake-up.

Description

Technical Field

[0001] The present disclosure relates to the field of voice processing technology, and in particular to a voice wake-up method, device, electronic device, medium and program product. Background Art

[0002] As the entry-dependent technology of intelligent voice interaction technology, voice wake-up technology is now used as the basic configuration technology of products by more and more electronic devices. The voice wake-up function of electronic devices is preset with a wake-up voice. When the current ambient sound including the wake-up voice is recognized, the voice wake-up function enters the wake-up state.

[0003] In order to provide users with a good user experience, it is necessary to avoid false wake-up or difficult wake-up of the voice wake-up function. However, due to factors such as the environment and similar voices, it is difficult to ensure the accuracy of the voice wake-up function. Summary of the invention

[0004] In order to overcome the problems existing in the related art, the present disclosure provides a voice wake-up method, device, electronic device, medium and program product. The technical solution of the present disclosure is as follows:

[0005] According to a first aspect of an embodiment of the present disclosure, a voice wake-up method is provided, including:

[0006] Get the current status of the voice wake-up function and the current ambient sound;

[0007] In the case where the current state is a dormant state, identifying whether the current ambient sound includes a wake-up voice, and obtaining a first-level recognition result;

[0008] When the first-level recognition result indicates that the current ambient sound includes a wake-up voice, sending the current ambient sound to a server;

[0009] Receive a second-level recognition result returned by the server, where the second-level recognition result is a result of the server identifying whether the current ambient sound meets a target wake-up condition;

[0010] When it is determined that the second-level recognition result indicates that the current ambient sound meets the target wake-up condition, the current state is switched from the sleep state to the wake-up state.

[0011] Optionally, the method further comprises:

[0012] When the current state is the awake state, performing overall speech enhancement on the speech features of the current ambient sound, wherein the overall speech enhancement represents enhancing the pure speech features in the speech features of the current ambient sound;

[0013] According to the speech features of the current ambient sound after overall speech enhancement, a response signal of the current ambient sound is acquired and output.

[0014] Optionally, when the current state is a dormant state, identifying whether the current ambient sound includes a wake-up voice includes:

[0015] When the current state is a dormant state, obtaining a voice feature of the current ambient sound;

[0016] According to the wake-up speech feature, performing wake-up speech enhancement on the speech feature of the current ambient sound, wherein the wake-up speech enhancement represents enhancing the wake-up speech feature in the speech feature of the current ambient sound;

[0017] According to the voice feature of the current ambient sound enhanced by the wake-up voice, it is identified whether the current ambient sound includes the wake-up voice.

[0018] Optionally, when the first-level recognition result indicates that the current ambient sound includes a wake-up voice, sending the current ambient sound to a server includes:

[0019] In a case where the first-level recognition result indicates that the current ambient sound includes a wake-up speech, the speech features of the current ambient sound enhanced by the wake-up speech are sent to a server.

[0020] Optionally, performing wake-up speech enhancement on the speech feature of the current ambient sound according to the wake-up speech feature includes:

[0021] Inputting the speech features of the current ambient sound into a model for learning the wake-up speech features to perform wake-up speech enhancement;

[0022] The training process of the model for learning the wake-up speech features includes at least the following steps:

[0023] Acquire a feature of a wake-up voice signal and a clean voice feature contained therein, and acquire a feature of a non-wake-up voice signal and a clean voice feature contained therein, wherein the clean voice feature of the wake-up voice signal contains the wake-up voice feature;

[0024] Inputting the feature of the wake-up voice signal, the feature of the non-wake-up voice signal, and the identifier of the dormant state into a preset model;

[0025] Obtaining a first estimated feature output by the preset model after performing wake-up speech enhancement on the feature of the wake-up speech signal, and obtaining a second estimated feature output by the preset model after performing wake-up speech enhancement on the feature of the non-wake-up speech signal;

[0026] generating a first loss function value according to the difference between the first estimated feature and the second estimated feature and the corresponding pure speech feature;

[0027] According to the first loss function value, the model parameters of the preset model are updated to obtain a model that learns the wake-up speech features.

[0028] Optionally, the training process of the model for learning the wake-up speech features further includes the following steps:

[0029] identifying the first estimated feature and outputting a first estimated state, and identifying the second estimated feature and outputting a second estimated state;

[0030] generating a second loss function value according to whether the first estimated state is an awake state and whether the second estimated state is a sleep state;

[0031] The method of updating the model parameters of the preset model according to the first loss function value to obtain a model that learns the wake-up speech feature includes:

[0032] According to the first loss function value and the second loss function value, the model parameters of the preset model are updated to obtain a model that learns the wake-up speech features.

[0033] Optionally, the training process of the model for learning the wake-up speech features further includes the following steps:

[0034] In the process of updating the model parameters of the preset model, the characteristics of the sample speech signal and the pure speech characteristics contained therein, as well as the identification of the awakening state are input into the preset model after the model parameters are updated;

[0035] Obtaining a third estimated feature outputted by the preset model after the model parameters are updated after overall speech enhancement is performed on the features of the sample speech signal;

[0036] Generating a third loss function value according to the difference between the third estimated feature and the pure speech feature;

[0037] According to the third loss function value, the preset model after the parameter update is updated to obtain a model that has learned the wake-up speech features, wherein the model that has learned the wake-up speech features is also used to perform overall speech enhancement on the speech features of the current ambient sound when the current state is the wake-up state, including: when the current state of the voice wake-up function is the wake-up state, the speech features of the current ambient sound are input into the model that has learned the wake-up speech features to perform overall speech enhancement.

[0038] Optionally, identifying whether the current ambient sound includes a wake-up voice includes:

[0039] Identify, by a pre-trained first wake-up module, whether the voice feature of the current ambient sound includes a wake-up voice feature, wherein the first wake-up module is trained using the wake-up voice as a training sample;

[0040] The second-level recognition result returned by the server is obtained by recognition through a pre-trained second wake-up module, where the second wake-up module is first trained with a general corpus as a training sample and then trained with a wake-up voice as a training sample;

[0041] Among them, the probability that the first wake-up module recognizes that the voice feature of the current ambient sound includes the wake-up voice feature is higher than the probability that the second wake-up module recognizes that the current ambient sound meets the target wake-up condition.

[0042] According to a second aspect of an embodiment of the present disclosure, a voice wake-up device is provided, including:

[0043] An acquisition module, configured to acquire the current state of the voice wake-up function and the current ambient sound;

[0044] a recognition module configured to, when the current state is a dormant state, identify whether the current ambient sound includes a wake-up voice, and obtain a first-level recognition result;

[0045] a sending module, configured to send the current ambient sound to a server when the first-level recognition result indicates that the current ambient sound includes a wake-up voice;

[0046] A receiving module is configured to receive a second-level recognition result returned by the server, where the second-level recognition result is a result of the server identifying whether the current ambient sound meets a target wake-up condition;

[0047] The state switching module is configured to switch the current state from the sleep state to the wake-up state when it is determined that the second-level recognition result indicates that the current ambient sound meets the target wake-up condition.

[0048] Optionally, the device further comprises:

[0049] An overall enhancement module is configured to perform overall speech enhancement on the speech features of the current ambient sound when the current state is the awake state, wherein the overall speech enhancement represents enhancement of pure speech features in the speech features of the current ambient sound;

[0050] The response module is configured to acquire and output a response signal of the current ambient sound according to the voice features of the current ambient sound after overall voice enhancement.

[0051] Optionally, the identification module includes:

[0052] an acquisition unit, configured to acquire the voice feature of the current ambient sound when the current state is a dormant state;

[0053] A wake-up speech enhancement unit is configured to perform wake-up speech enhancement on the speech feature of the current ambient sound according to the wake-up speech feature, wherein the wake-up speech enhancement represents enhancing the wake-up speech feature in the speech feature of the current ambient sound;

[0054] The recognition unit is configured to recognize whether the current ambient sound includes the wake-up speech according to the speech feature of the current ambient sound enhanced by the wake-up speech.

[0055] Optionally, the sending module includes:

[0056] The sending unit is configured to send the voice features of the current ambient sound enhanced by the wake-up speech to the server when the first-level recognition result indicates that the current ambient sound includes the wake-up speech.

[0057] Optionally, the wake-up voice enhancement unit includes:

[0058] A wake-up speech enhancement subunit is configured to input the speech feature of the current ambient sound into a model for learning the wake-up speech feature, so as to perform wake-up speech enhancement;

[0059] The training process of the model for learning the wake-up speech features includes at least the following steps:

[0060] Acquire a feature of a wake-up voice signal and a clean voice feature contained therein, and acquire a feature of a non-wake-up voice signal and a clean voice feature contained therein, wherein the clean voice feature of the wake-up voice signal contains the wake-up voice feature;

[0061] Inputting the feature of the wake-up voice signal, the feature of the non-wake-up voice signal, and the identifier of the dormant state into a preset model;

[0062] Obtaining a first estimated feature output by the preset model after performing wake-up speech enhancement on the feature of the wake-up speech signal, and obtaining a second estimated feature output by the preset model after performing wake-up speech enhancement on the feature of the non-wake-up speech signal;

[0063] generating a first loss function value according to the difference between the first estimated feature and the second estimated feature and the corresponding pure speech feature;

[0064] According to the first loss function value, the model parameters of the preset model are updated to obtain a model that learns the wake-up speech features.

[0065] Optionally, the training process of the model for learning the wake-up speech features further includes the following steps:

[0066] identifying the first estimated feature and outputting a first estimated state, and identifying the second estimated feature and outputting a second estimated state;

[0067] generating a second loss function value according to whether the first estimated state is an awake state and whether the second estimated state is a sleep state;

[0068] The method of updating the model parameters of the preset model according to the first loss function value to obtain a model that learns the wake-up speech feature includes:

[0069] According to the first loss function value and the second loss function value, the model parameters of the preset model are updated to obtain a model that learns the wake-up speech features.

[0070] Optionally, the training process of the model for learning the wake-up speech features further includes the following steps:

[0071] In the process of updating the model parameters of the preset model, the characteristics of the sample speech signal and the pure speech characteristics contained therein, as well as the identification of the awakening state are input into the preset model after the model parameters are updated;

[0072] Obtaining a third estimated feature outputted by the preset model after the model parameters are updated after overall speech enhancement is performed on the features of the sample speech signal;

[0073] Generating a third loss function value according to the difference between the third estimated feature and the pure speech feature;

[0074] According to the third loss function value, the preset model after the parameter update is updated to obtain a model that has learned the wake-up speech features, wherein the model that has learned the wake-up speech features is also used to perform overall speech enhancement on the speech features of the current ambient sound when the current state is the wake-up state, including: when the current state of the voice wake-up function is the wake-up state, the speech features of the current ambient sound are input into the model that has learned the wake-up speech features to perform overall speech enhancement.

[0075] Optionally, the identification module includes:

[0076] A first recognition unit is configured to recognize whether the voice feature of the current ambient sound includes a wake-up voice feature through a pre-trained first wake-up module, wherein the first wake-up module is trained using the wake-up voice as a training sample;

[0077] The second-level recognition result returned by the server and received by the receiving module is obtained by recognition by a pre-trained second wake-up module, wherein the second wake-up module is first trained with a general corpus as a training sample and then trained with a wake-up speech as a training sample;

[0078] Among them, the probability that the first wake-up module recognizes that the voice feature of the current ambient sound includes the wake-up voice feature is higher than the probability that the second wake-up module recognizes that the current ambient sound meets the target wake-up condition.

[0079] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, including:

[0080] processor;

[0081] a memory for storing instructions executable by the processor;

[0082] The processor is configured to execute the instructions to implement the voice wake-up method as described in the first aspect.

[0083] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the voice wake-up method as described in the first aspect.

[0084] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, including a computer program or a computer instruction, wherein when the computer program or the computer instruction is executed by a processor, the voice wake-up method as described in the first aspect is implemented.

[0085] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects:

[0086] In the present disclosure, the first level recognition result of identifying whether the current ambient sound includes the wake-up voice is first obtained. If the first level recognition result indicates that the current ambient sound includes the wake-up voice, the second level recognition result of the server identifying whether the current ambient sound meets the target wake-up condition needs to be received. If the second level recognition result indicates that the current ambient sound meets the target wake-up condition, the current state will be switched from the dormant state to the wake-up state. The first level recognition result and the second level recognition result are combined to determine whether to switch the state of the voice wake-up function to the wake-up state, ensuring a high accuracy rate of voice wake-up.

[0087] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0089] Figure 1 is a flow chart of a voice wake-up method according to an exemplary embodiment;

[0090] Figure 2 is a processing block diagram of the voice wake-up method in the present disclosure;

[0091] Figure 3 is a schematic block diagram of a preset model in the present disclosure;

[0092] Figure 4 is a schematic block diagram of a model for learning wake-up speech features in the present disclosure;

[0093] Figure 5 is a block diagram of a voice wake-up device according to an exemplary embodiment;

[0094] Figure 6 The figure is a block diagram of a voice wake-up device according to an exemplary embodiment. DETAILED DESCRIPTION

[0095] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0096] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0097] Figure 1 is a flow chart of a voice wake-up method according to an exemplary embodiment. Figure 1 As shown, the voice wake-up method can be used in electronic devices such as computers, mobile phones, and tablet computers, and includes the following steps.

[0098] In step S11, the current state of the voice wake-up function and the current ambient sound are obtained.

[0099] The current state of the voice wake-up function includes the wake-up state or the sleep state. The current ambient sound is the audio in the current environment monitored in real time through an audio collection device such as a handset or microphone.

[0100] In step S12, when the current state is the dormant state, it is identified whether the current ambient sound includes a wake-up voice to obtain a first-level recognition result.

[0101] The voice wake-up function has preset the wake-up voice in advance. When the current state is the sleep state, it is identified whether the current ambient sound obtained includes the wake-up voice to obtain the first-level recognition result.

[0102] In order to improve the recognition accuracy, the current ambient sound can be processed. Optionally, the collected current ambient sound can be processed with linear echo cancellation to obtain the current ambient sound with some echoes eliminated or weakened. When collecting the current ambient sound, the voice signal played by the device itself is obtained at the same time, and the collected current ambient sound is processed with linear echo cancellation using the voice signal played by the device itself. Specifically, the processing can be performed using the following formula:

[0103] res(n)=LinearAEC[rgf(n),d(n)]

[0104] Wherein, ref(n) is the voice signal played by the device itself; d(n) is the current ambient sound collected; res(n) is the current ambient sound processed with linear echo cancellation, and Linera AEC[*,#] represents the method for implementing linear echo cancellation, including the echo cancellation method based on the block frequency domain adaptive algorithm, the echo cancellation method based on subband decomposition and minimum mean square error, etc. The present disclosure does not limit the method for implementing linear echo cancellation. In addition, noise elimination can also be performed on the current ambient sound.

[0105] Optionally, identifying whether the current ambient sound includes the wake-up voice may be identifying whether the voice features of the current ambient sound include the wake-up voice features, where the voice features include: Mel spectrum, energy features, short-time zero-crossing rate, fundamental frequency, harmonics, time domain characteristics, frequency domain characteristics, etc. The first recognition result is the result of the electronic device end recognizing the current ambient sound.

[0106] The processed current ambient sound is subjected to feature extraction to obtain the features of the current ambient sound, including: inputting ref(n) and res(n) into a feature extraction module, performing feature transformation on ref(n) and res(n) respectively, and combining the features of the transformed ref(n) and res(n), wherein the feature transformation method may be a power spectrum, a Mel cepstrum, a Bark cepstrum, etc., and the feature combination method may be a method based on frequency band interval splicing or overall splicing, etc., which is not limited in the present disclosure. According to the wake-up speech, the wake-up speech features are obtained.

[0107] If the voice features of the current ambient sound include the wake-up voice features, the first-level recognition result of identifying whether the current ambient sound includes the wake-up voice is included; if the voice features of the current ambient sound do not include the wake-up voice features, the first-level recognition result of identifying whether the current ambient sound includes the wake-up voice is not included.

[0108] Among them, identifying whether the current ambient sound includes the wake-up voice is the probability that the electronic device recognizes that the current ambient sound includes the wake-up voice. When the probability recognized by the electronic device is higher than the first preset probability, the first-level recognition result indicates that the current ambient sound includes the wake-up voice.

[0109] In step S13, when the first-level recognition result indicates that the current ambient sound includes a wake-up voice, the current ambient sound is sent to a server.

[0110] When the current state is the sleep state, in order to more accurately determine whether to wake up and avoid false wake-up, if the first-level recognition result indicates that the current ambient sound includes the wake-up voice, the current ambient sound needs to be sent to the server for secondary recognition to confirm whether to switch the current state from the sleep state to the wake-up state.

[0111] Optionally, the current environment voice sent to the server is the current environment sound with echo and noise eliminated.

[0112] Optionally, when the first-level recognition result indicates that the current ambient sound includes a wake-up voice, the current state may be switched from the sleep state to the pre-wake-up state.

[0113] In step S14, a second-level recognition result returned by the server is received, where the second-level recognition result is a result of the server identifying whether the current ambient sound meets the target wake-up condition.

[0114] The current ambient sound is sent to the server, and the server or a module contained in the server can perform a second recognition on the current ambient sound to obtain a second-level recognition result of whether the current ambient sound meets the target wake-up condition, and return the second-level recognition result.

[0115] The target wake-up condition is that the current ambient sound includes the wake-up voice. Identifying whether the current ambient sound meets the target wake-up condition (whether the current ambient sound includes the wake-up voice) is the probability that the server recognizes that the current ambient sound includes the wake-up voice. When the probability recognized by the server is higher than the second preset probability, the second-level recognition result indicates that the current ambient sound includes the wake-up voice. Among them, for the same voice, the probability that the server recognizes that it includes the wake-up voice is lower than the probability that the electronic device recognizes that it includes the wake-up voice.

[0116] In step S15, when it is determined that the second-level recognition result indicates that the current ambient sound meets the target wake-up condition, the current state is switched from the sleep state to the wake-up state.

[0117] If the second-level recognition result indicates that the current ambient sound does not meet the target wake-up condition, the first-level recognition result is considered to be a misrecognition, so the current state is kept in the dormant state. Optionally, if the first-level recognition result indicates that the current ambient sound includes a wake-up voice, the current state is switched from the dormant state to the pre-wake-up state, and if the second-level recognition result indicates that the current ambient sound does not meet the target wake-up condition, the current state is switched from the pre-wake-up state back to the dormant state.

[0118] When the second-level recognition result indicates that the current ambient sound meets the target wake-up condition, the first-level recognition result is considered to be a correct recognition result, so the current state is switched to the wake-up state. When the current state is the wake-up state, a response signal of the current ambient sound is obtained and output.

[0119] In the present disclosure, the first level recognition result of identifying whether the current ambient sound includes the wake-up voice is first obtained. If the first level recognition result indicates that the current ambient sound includes the wake-up voice, the second level recognition result of the server identifying whether the current ambient sound meets the target wake-up condition needs to be received. If the second level recognition result indicates that the current ambient sound meets the target wake-up condition, the current state will be switched from the dormant state to the wake-up state. The first level recognition result and the second level recognition result are combined to determine whether to switch the state of the voice wake-up function to the wake-up state, ensuring a high accuracy rate of voice wake-up.

[0120] Figure 2 This is a processing block diagram of the voice wake-up method, such as Figure 2 As shown, the voice wake-up method can make a second judgment after making a judgment on whether to wake up the current ambient sound.

[0121] The first wake-up module is a module on the electronic device. The electronic device monitors the current ambient sound in real time with low power consumption, and uses the first wake-up module to identify whether the current ambient sound includes the wake-up voice, and determines whether to wake up according to the recognition result of the first wake-up module. The first wake-up module has a high probability of identifying that the current ambient sound includes the wake-up voice to avoid missing the wake-up.

[0122] The second wake-up module is a module on the server. The second wake-up module identifies whether the current ambient sound includes the wake-up voice, and determines whether to wake up according to the recognition result of the second wake-up module. The probability that the second wake-up module recognizes that the current ambient sound includes the wake-up voice is lower than the probability that the first wake-up module recognizes, so as to avoid false wake-up.

[0123] In the related art, the voice wake-up function of the electronic device recognizes the current ambient sound in real time with low power consumption, but the recognition accuracy of this method is not high. In order to ensure the accuracy of identifying whether the current ambient sound includes the wake-up voice, so as to avoid false wake-up or difficult wake-up, the voice wake-up function of the electronic device needs to recognize the current ambient sound in real time with higher power consumption, and this method requires a lot of power consumption for the electronic device.

[0124] In the present disclosure, the accuracy of the wake-up function can be ensured by combining a first wake-up module with a high recognition probability and a second wake-up module with a low recognition probability.

[0125] On the basis of the above technical solution, the voice wake-up method also includes: when the current state is the wake-up state, performing overall voice enhancement on the voice features of the current ambient sound, wherein the overall voice enhancement characterizes the enhancement of the pure voice features in the voice features of the current ambient sound; and according to the voice features of the current ambient sound after the overall voice enhancement, obtaining and outputting a response signal of the current ambient sound.

[0126] When the current state of the voice wake-up function is the wake-up state, there is no need to repeatedly determine whether to wake up, so there is no need to determine whether the current ambient sound includes the wake-up voice. At this time, a response signal of the current ambient sound can be obtained and output according to the current ambient sound, so as to perform the corresponding task according to the response signal.

[0127] In order to more accurately determine the task corresponding to the current ambient sound to be executed, the current ambient sound may be subjected to overall speech enhancement, which refers to removing or weakening echoes, noises, etc. contained in the speech signal while retaining the pure speech.

[0128] The linear echo cancellation processing of the current ambient sound introduced in the previous article is essentially a method of overall speech enhancement. The following article also introduces a method of overall speech enhancement of the current ambient sound by learning a model of wake-up speech features.

[0129] The voice wake-up method provided by this exemplary embodiment can, when the current state of the voice wake-up function is the wake-up state, perform overall voice enhancement on the current ambient sound to remove or weaken echoes, noise, etc. contained in the voice signal while retaining pure voice; in this way, based on the current ambient sound after the overall voice enhancement, the response signal of the current ambient sound can be more accurately acquired and output.

[0130] On the basis of the above technical solution, when the current state is a sleep state, identifying whether the current ambient sound includes a wake-up speech includes: when the current state is a sleep state, acquiring speech features of the current ambient sound; performing wake-up speech enhancement on the speech features of the current ambient sound according to the wake-up speech features, wherein the wake-up speech enhancement characterizes the enhancement of the wake-up speech features in the speech features of the current ambient sound; and identifying whether the current ambient sound includes a wake-up speech according to the speech features of the current ambient sound enhanced by the wake-up speech.

[0131] When the current state is the dormant state, in order to improve the accuracy of identifying whether the current ambient sound includes the wake-up voice, the current ambient sound may be firstly enhanced for the wake-up voice, and then the recognition is performed based on the current ambient sound after the wake-up voice enhancement.

[0132] Wake-up speech enhancement refers to removing or weakening the echo and noise features in the speech features of the current ambient sound, retaining the features of the pure speech, and especially focusing on retaining the wake-up speech features in the pure speech features. The wake-up speech feature is the feature of the wake-up speech. Wake-up speech enhancement is a special case of overall speech enhancement. On the basis of overall speech enhancement, it focuses more on retaining the wake-up speech features in the pure speech features to highlight the wake-up speech features. The following article introduces a method for performing wake-up speech enhancement on the current ambient sound by learning a model with wake-up speech features.

[0133] Identifying whether the current ambient sound includes the wake-up voice may be achieved by identifying whether the voice feature of the current ambient sound includes the wake-up voice feature.

[0134] Optionally, when the first-level recognition result indicates that the current ambient sound includes the wake-up speech, the current ambient sound sent to the server may also be the current ambient sound after the wake-up speech is enhanced.

[0135] The voice wake-up method provided by this exemplary embodiment, when identifying whether the current ambient sound includes the wake-up voice, enhances the current ambient sound by the wake-up voice, and recognizes the current ambient sound after the wake-up voice enhancement, thereby improving the recognition accuracy.

[0136] Based on the above technical solution, the method for performing wake-up speech enhancement on the current ambient sound may be to input the speech features of the current ambient sound into a model that has learned the wake-up speech features, so as to obtain the current ambient sound after wake-up speech enhancement output by the model that has learned the wake-up speech features.

[0137] The model that has learned the wake-up speech features can be trained through the following steps: obtaining the features of the wake-up speech signal and the pure speech features contained therein, and obtaining the features of the non-wake-up speech signal and the pure speech features contained therein, wherein the pure speech features of the wake-up speech signal include the wake-up speech features; inputting the features of the wake-up speech signal, the features of the non-wake-up speech signal, and the identifier of the sleep state into a preset model; obtaining the first estimated features output by the preset model after performing wake-up speech enhancement on the features of the wake-up speech signal, and obtaining the second estimated features output by the preset model after performing wake-up speech enhancement on the features of the non-wake-up speech signal; generating a first loss function value based on the differences between the first estimated features and the second estimated features and the corresponding pure speech features; and updating the model parameters of the preset model based on the first loss function value to obtain a model that has learned the wake-up speech features.

[0138] Figure 3 is a schematic diagram of the preset model, such as Figure 3 As shown, the preset model includes an enhancement module, an embedding module, a recognition module and a loss function module.

[0139] In order to train the preset model, it is necessary to obtain training samples first. Among them, the positive sample is the feature of the wake-up speech signal, which includes the wake-up speech feature, echo feature, noise feature and pure speech feature of the non-wake-up speech feature, wherein the wake-up speech feature and the pure speech feature of the non-wake-up speech feature constitute the pure speech feature; the negative sample is the feature of the non-wake-up speech signal, which includes the pure speech feature, echo feature and noise feature of the non-wake-up speech feature. Among them, the positive sample and the negative sample are obtained by performing linear echo elimination processing, feature transformation, feature combination, etc. on the speech containing the wake-up speech and the speech not containing the wake-up speech, respectively, wherein the method of linear echo elimination processing, feature transformation and feature combination can refer to the above description.

[0140] The features of the wake-up speech signal and the features of the non-wake-up speech signal are input into the enhancement module of the preset model, and the identification of the dormant state is input into the enhancement module of the preset model through the embedding module. During training, the state identification input into the enhancement module by the embedding module is preset; when using the trained model, the state identification input into the enhancement module by the embedding module is obtained according to the recognition result of the recognition module.

[0141] The enhancement module determines to perform wake-up speech enhancement processing on the input features according to the input sleep state identifier, obtains the first estimated feature output after the wake-up speech enhancement is performed on the features of the wake-up speech signal, and the second estimated feature output after the wake-up speech enhancement is performed on the features of the non-wake-up speech signal. Because the features of the non-wake-up speech signal do not include the wake-up word feature, the wake-up speech enhancement is performed on the features of the non-wake-up speech signal, and the second estimated feature obtained is a feature in which all features are weakened.

[0142] The preset model can process the features of the speech signal using one or more of a convolutional neural network, a long short-term memory network, a gated recurrent unit network, a fully connected network, and other networks, and the present disclosure does not limit this.

[0143] The loss module generates a first loss function value according to the difference between the first estimated feature and its corresponding clean speech feature, and the difference between the second estimated feature and its corresponding clean speech feature; wherein the first loss function value can be calculated using a variety of specific loss functions such as signal-to-noise ratio, scale-invariant signal-to-noise ratio and its variants, mean square error, or a combination of a variety of loss functions.

[0144] The loss module updates the model parameters of the preset model through the first loss function value, and guides the enhancement module to better perform wake-up speech enhancement to obtain a model that learns the wake-up word features.

[0145] Based on the above technical solution, the training process of the model that has learned the wake-up speech features also includes the following steps: identifying the first estimated feature and outputting a first estimated state, and identifying the second estimated feature and outputting a second estimated state; generating a second loss function value based on whether the first estimated state is a wake-up state and whether the second estimated state is a sleep state; updating the model parameters of the preset model based on the first loss function value to obtain the model that has learned the wake-up speech features, including: updating the model parameters of the preset model based on the first loss function value and the second loss function value to obtain the model that has learned the wake-up speech features.

[0146] One of the purposes of obtaining a model that learns the wake-up speech feature is to perform wake-up speech enhancement on the features of the current ambient sound, and the role of the features of the current ambient sound after the wake-up speech enhancement is to be identified to determine whether it contains the wake-up speech feature. Therefore, in order to make the features of the current ambient sound after the wake-up speech enhancement more accurately identified, when training the preset model, the recognition module is used to identify the features of the current ambient sound after the wake-up speech enhancement to identify whether it contains the wake-up speech feature. Then, according to the recognition result output by the recognition module and whether the current ambient sound features corresponding to the recognition result actually contain the wake-up speech feature, a second loss function value is established.

[0147] The first estimated state is obtained by identifying based on the first estimated feature, and the first estimated feature is the feature obtained by performing wake-up speech enhancement on the feature of the wake-up speech signal, and the feature includes the wake-up speech feature; the second estimated state is obtained by identifying based on the second estimated feature, and the second estimated feature is the feature obtained by performing wake-up speech enhancement on the feature of the non-wake-up speech signal, and the feature does not include the wake-up word feature. Therefore, the first estimated feature should actually correspond to the wake-up state, and the second estimated feature should actually correspond to the sleep state. Therefore, a second loss function value can be generated based on whether the first estimated state is the wake-up state and whether the second estimated state is the sleep state, and the second loss function value can be represented by a loss function such as cross entropy.

[0148] According to the combined effect of the first loss function value and the second loss function value, the model parameters of the preset model are updated to obtain a model that has learned the wake-up speech features. Wherein, weight parameters can be set for the first loss function value and the second loss function value respectively, and when the model parameters of the preset model are updated, the first loss function value and the second loss function value act according to the corresponding weight parameters.

[0149] On the basis of the above technical scheme, the training process of the model that learns the wake-up speech features also includes the following steps: in the process of updating the model parameters of the preset model, the features of the sample speech signal and the pure speech features it contains, as well as the identification of the wake-up state are input into the preset model after the model parameters are updated; the third estimated features output by the preset model after the model parameters are updated are obtained after the overall speech enhancement is performed on the features of the sample speech signal; a third loss function value is generated based on the difference between the third estimated features and the pure speech features; based on the third loss function value, the preset model after the parameters are updated is updated to obtain the model that learns the wake-up speech features.

[0150] When the current state of the voice wake-up function is the wake-up state or the sleep state, the characteristics of the voice signal need to be purified. Therefore, the inventor hopes to use a model to achieve both overall voice enhancement of the voice signal in the wake-up state and wake-up voice enhancement of the voice signal in the sleep state. To this end, the training of the preset model may also include:

[0151] In the process of updating the model parameters of the preset model, the characteristics of the sample speech signal and the pure speech characteristics it contains, as well as the identification of the wake-up state are input into the preset model after the model parameters are updated, wherein the sample speech signal does not contain the wake-up speech signal, and the characteristics of the non-wake-up speech signal can be directly used as the characteristics of the sample speech signal.

[0152] When the embedding module inputs the identifier of the awakening state to the enhancement module, the enhancement module performs overall speech enhancement on the features of the sample speech signal and outputs a third estimated feature. The loss module generates a third loss function value based on the difference between the third estimated feature and the pure speech feature. The third loss function value can be calculated using a variety of specific loss functions such as signal-to-noise ratio, scale-invariant signal-to-noise ratio and its variants, mean square error, or a combination of multiple loss functions. According to the third loss function value, the preset model after parameter update is updated to obtain a pre-trained model. When the current awakening state is the awakening state, the features of the speech signal are input into the pre-trained model for overall speech enhancement.

[0153] In this way, the model that has learned the wake-up speech features can be used to perform overall speech enhancement on the speech features of the current ambient sound.

[0154] Figure 4 is a schematic diagram of a model that learns the wake-up speech features, such as Figure 4 As shown, the model for learning the wake-up speech features includes an enhancement module, an embedding module and a recognition module.

[0155] The enhancement module in the model that has learned the wake-up speech feature can determine whether to perform wake-up speech enhancement or overall speech enhancement on the features of the input speech signal according to the wake-up state identifier or the sleep state identifier input by the embedding module. In the sleep state, the features of the speech signal after the wake-up speech enhancement are input to the recognition module, and the recognition module identifies whether the features of the speech signal after the wake-up speech enhancement contain the wake-up speech feature, and outputs the wake-up state when the features of the speech signal after the wake-up speech enhancement are identified to contain the wake-up speech feature, and outputs the sleep state when the features of the speech signal after the wake-up speech enhancement are identified to not contain the wake-up speech feature. The wake-up state or sleep state output by the recognition module is sent to the embedding module, and the embedding module generates a corresponding state identifier according to the received wake-up state or sleep state, and inputs it into the enhancement module.

[0156] When the wake-up state identifier is input from the embedding module to the enhancement module, there is no need to further determine whether the speech signal feature contains the wake-up speech feature at this time. Therefore, the enhancement module performs overall speech enhancement on the speech signal feature. At this time, the enhancement module may not send the overall purified speech signal feature to the recognition module, but send it to the corresponding place. In practical applications, according to different requirements, the recognition module can be changed, or different recognition probabilities can be set for the recognition module.

[0157] Based on the above technical solution, identifying whether the current ambient sound includes a wake-up speech includes: identifying whether the speech feature of the current ambient sound contains a wake-up speech feature through a pre-trained first wake-up module, where the first wake-up module is trained with the wake-up speech as a training sample; the second-level recognition result returned by the server is obtained by identifying through a pre-trained second wake-up module, and the second wake-up module is first trained with a general corpus as a training sample and then trained in combination with the wake-up speech as a training sample; wherein, the probability that the first wake-up module identifies that the speech feature of the current ambient sound contains a wake-up speech feature is higher than the probability that the second wake-up module identifies that the current ambient sound meets the target wake-up condition.

[0158] The first wake-up module identifies whether the speech feature of the current ambient sound contains a wake-up speech feature, and the first wake-up module is trained with the wake-up speech as a training sample. The first wake-up module can be a module on the electronic device side. Considering the limited computing power and resources, the first wake-up module cannot be too complex, but it is necessary to ensure the recall of the wake-up speech. To improve the wake-up rate and be more robust to different voices, the training samples of the first wake-up module are a large amount of speech data containing wake-up speech, and the wake-up speech is a non-tonal syllable. For example, the wake-up speech is "car", corresponding to "xiao che", rather than directly using the corresponding characters.

[0159] The second-level recognition result returned by the server can be to identify whether the current ambient sound contains a wake-up speech through a pre-trained second wake-up module. The second wake-up module is first trained with the speech features of the general corpus as a training sample and then trained in combination with the speech containing the wake-up speech as a training sample. The second wake-up module is trained with the features of a large number of general speech signals. At the same time, it can also use speech signals outside the large-scale wake-up speech. Then, it uses the corpus with wake-up speech, plus a small part of the general corpus, and trains on the basis of the general model to obtain a second wake-up module with high recognition accuracy for wake-up speech.

[0160] The first wake-up module continuously monitors the surrounding environment with low power consumption, receives the current ambient sound in the surrounding environment, and when it is recognized that the current ambient sound includes the wake-up voice, switches the current state from the sleep state to the pre-wake-up state, and sends the current ambient sound to the server. The second wake-up module on the server performs secondary recognition. If the second wake-up module recognizes that the current ambient sound does not include the wake-up voice, it switches the current state from the pre-wake-up state to the sleep state. If it recognizes that the current ambient sound includes the wake-up voice, it switches the current state from the pre-wake-up state to the wake-up state, and continuously receives the current ambient sound for subsequent operations.

[0161] In order to ensure timely response, the first wake-up module can be set to have a higher recognition probability. As long as the probability of recognizing that the wake-up word feature is included in the voice signal feature is higher than the first preset probability, the wake-up state is output. The secondary recognition is to ensure the accuracy of the wake-up and avoid misidentification and false wake-up. Therefore, the second preset probability of recognition of the second wake-up module can be set lower than the first preset probability of recognition of the first wake-up module. The higher the recognition probability, the easier it is to wake up.

[0162] The voice wake-up method provided by this exemplary embodiment can ensure that the electronic device monitors the surrounding environment with lower power consumption and responds to the user's wake-up voice in a timely manner; and sets the server to recognize the voice signal features with a lower recognition probability to avoid false wake-ups.

[0163] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

[0164] Figure 5 is a block diagram of a voice wake-up device according to an exemplary embodiment. Figure 5 The device includes an acquisition module 51, an identification module 52, a sending module 53, a receiving module 54 and a state switching module 55.

[0165] The acquisition module 51 is configured to acquire the current state of the voice wake-up function and the current ambient sound;

[0166] The recognition module 52 is configured to, when the current state is the dormant state, identify whether the current ambient sound includes a wake-up voice, and obtain a first-level recognition result;

[0167] The sending module 53 is configured to send the current ambient sound to the server when the first-level recognition result indicates that the current ambient sound includes the wake-up voice;

[0168] The receiving module 54 is configured to receive a second-level recognition result returned by the server, where the second-level recognition result is a result of the server identifying whether the current ambient sound meets the target wake-up condition;

[0169] The state switching module 55 is configured to switch the current state from the sleep state to the wake-up state when it is determined that the second-level recognition result indicates that the current ambient sound meets the target wake-up condition.

[0170] Optionally, the device further comprises:

[0171] An overall enhancement module is configured to perform overall speech enhancement on the speech features of the current ambient sound when the current state is the awake state, wherein the overall speech enhancement represents enhancement of pure speech features in the speech features of the current ambient sound;

[0172] The response module is configured to acquire and output a response signal of the current ambient sound according to the voice features of the current ambient sound after overall voice enhancement.

[0173] Optionally, the identification module 52 includes:

[0174] an acquisition unit, configured to acquire the voice feature of the current ambient sound when the current state is a dormant state;

[0175] A wake-up speech enhancement unit is configured to perform wake-up speech enhancement on the speech feature of the current ambient sound according to the wake-up speech feature, wherein the wake-up speech enhancement represents enhancing the wake-up speech feature in the speech feature of the current ambient sound;

[0176] The recognition unit is configured to recognize whether the current ambient sound includes the wake-up speech according to the speech feature of the current ambient sound enhanced by the wake-up speech.

[0177] Optionally, the sending module 53 includes:

[0178] The sending unit is configured to send the voice features of the current ambient sound enhanced by the wake-up speech to the server when the first-level recognition result indicates that the current ambient sound includes the wake-up speech.

[0179] Optionally, the wake-up voice enhancement unit includes:

[0180] A wake-up speech enhancement subunit is configured to input the speech feature of the current ambient sound into a model for learning the wake-up speech feature, so as to perform wake-up speech enhancement;

[0181] The training process of the model for learning the wake-up speech features includes at least the following steps:

[0182] Acquire a feature of a wake-up voice signal and a clean voice feature contained therein, and acquire a feature of a non-wake-up voice signal and a clean voice feature contained therein, wherein the clean voice feature of the wake-up voice signal contains the wake-up voice feature;

[0183] Inputting the feature of the wake-up voice signal, the feature of the non-wake-up voice signal, and the identifier of the dormant state into a preset model;

[0184] Obtaining a first estimated feature output by the preset model after performing wake-up speech enhancement on the feature of the wake-up speech signal, and obtaining a second estimated feature output by the preset model after performing wake-up speech enhancement on the feature of the non-wake-up speech signal;

[0185] generating a first loss function value according to the difference between the first estimated feature and the second estimated feature and the corresponding pure speech feature;

[0186] According to the first loss function value, the model parameters of the preset model are updated to obtain a model that learns the wake-up speech features.

[0187] Optionally, the training process of the model for learning the wake-up speech features further includes the following steps:

[0188] identifying the first estimated feature and outputting a first estimated state, and identifying the second estimated feature and outputting a second estimated state;

[0189] generating a second loss function value according to whether the first estimated state is an awake state and whether the second estimated state is a sleep state;

[0190] The method of updating the model parameters of the preset model according to the first loss function value to obtain a model that learns the wake-up speech feature includes:

[0191] According to the first loss function value and the second loss function value, the model parameters of the preset model are updated to obtain a model that learns the wake-up speech features.

[0192] Optionally, the training process of the model for learning the wake-up speech features further includes the following steps:

[0193] In the process of updating the model parameters of the preset model, the characteristics of the sample speech signal and the pure speech characteristics contained therein, as well as the identification of the awakening state are input into the preset model after the model parameters are updated;

[0194] Obtaining a third estimated feature outputted by the preset model after the model parameters are updated after overall speech enhancement is performed on the features of the sample speech signal;

[0195] Generating a third loss function value according to the difference between the third estimated feature and the pure speech feature;

[0196] According to the third loss function value, the preset model after the parameter update is updated to obtain a model that has learned the wake-up speech features, wherein the model that has learned the wake-up speech features is also used to perform overall speech enhancement on the speech features of the current ambient sound when the current state is the wake-up state, including: when the current state of the voice wake-up function is the wake-up state, the speech features of the current ambient sound are input into the model that has learned the wake-up speech features to perform overall speech enhancement.

[0197] Optionally, the identification module 52 includes:

[0198] A first recognition unit is configured to recognize whether the voice feature of the current ambient sound includes a wake-up voice feature through a pre-trained first wake-up module, wherein the first wake-up module is trained using the wake-up voice as a training sample;

[0199] The second-level recognition result returned by the server and received by the receiving module is obtained by recognition by a pre-trained second wake-up module, wherein the second wake-up module is first trained with a general corpus as a training sample and then trained with a wake-up speech as a training sample;

[0200] Among them, the probability that the first wake-up module recognizes that the voice feature of the current ambient sound includes the wake-up voice feature is higher than the probability that the second wake-up module recognizes that the current ambient sound meets the target wake-up condition.

[0201] Figure 6 1 is a block diagram of a device 600 for voice wake-up according to an exemplary embodiment. For example, the device 600 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0202] Reference Figure 6 , the electronic device 600 may include one or more of the following components: a processing component 602 , a memory 604 , a power component 606 , a multimedia component 608 , an audio component 610 , an input / output (I / O) interface 612 , a sensor component 614 , and a communication component 616 .

[0203] The processing component 602 generally controls the overall operation of the device 600, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 602 may include one or more processors 620 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 602 may include one or more modules to facilitate the interaction between the processing component 602 and other components. For example, the processing component 602 may include a multimedia module to facilitate the interaction between the multimedia component 608 and the processing component 602.

[0204] The memory 604 is configured to store various types of data to support operations on the device 600. Examples of such data include instructions for any application or method operating on the device 600, contact data, phone book data, messages, pictures, videos, etc. The memory 604 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0205] The power supply component 606 provides power to the various components of the device 600. The power supply component 606 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device 600.

[0206] The multimedia component 608 includes a screen that provides an output interface between the device 600 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 608 includes a front camera and / or a rear camera. When the device 600 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0207] The audio component 610 is configured to output and / or input audio signals. For example, the audio component 610 includes a microphone (MIC), and when the device 600 is in an operating mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 604 or sent via the communication component 616. In some embodiments, the audio component 610 also includes a speaker for outputting audio signals.

[0208] I / O interface 612 provides an interface between processing component 602 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: a home button, a volume button, a start button, and a lock button.

[0209] The sensor assembly 614 includes one or more sensors for providing various aspects of status assessment for the device 600. For example, the sensor assembly 614 can detect the open / closed state of the device 600, the relative positioning of components, such as the display and keypad of the device 600, and the sensor assembly 614 can also detect the position change of the device 600 or a component of the device 600, the presence or absence of user contact with the device 600, the orientation or acceleration / deceleration of the device 600, and the temperature change of the device 600. The sensor assembly 614 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 614 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 614 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0210] The communication component 616 is configured to facilitate wired or wireless communication between the device 600 and other devices. The device 600 can access a wireless network based on a communication standard, such as WiFi, a carrier network (such as 2G, 3G, 4G or 5G), or a combination thereof. In an exemplary embodiment, the communication component 616 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 616 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0211] In an exemplary embodiment, the electronic device 600 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above-mentioned voice wake-up method.

[0212] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 604 including instructions, and the instructions can be executed by a processor 620 of the device 600 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0213] In an exemplary embodiment, a computer program product is also provided, the computer program product includes a readable program code, the readable program code can be executed by the processor 620 of the electronic device 600 to complete the above-mentioned personnel authority management method. Optionally, the program code can be stored in a storage medium of the electronic device 600, and the storage medium can be a non-temporary computer-readable storage medium, for example, the non-temporary computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0214] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application 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 in this disclosure. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are 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.

Claims

1. A voice wake-up method, characterized in that: The method comprises: Get the current status of the voice wake-up function and the current ambient sound; In the case where the current state is a dormant state, identifying whether the current ambient sound includes a wake-up voice, and obtaining a first-level recognition result; When the first-level recognition result indicates that the current ambient sound includes a wake-up voice, sending the current ambient sound to a server; Receive a second-level recognition result returned by the server, where the second-level recognition result is a result of the server identifying whether the current ambient sound meets a target wake-up condition; If it is determined that the second-level recognition result indicates that the current ambient sound meets the target wake-up condition, switching the current state from the dormant state to the awake state; In the case where the current state is a dormant state, the method further includes: inputting a speech feature of the current ambient sound into a model that has learned a wake-up speech feature, obtaining the current ambient sound after wake-up speech enhancement output by the model that has learned the wake-up speech feature, wherein the wake-up speech enhancement characterizes the enhancement of the wake-up speech feature in the speech feature of the current ambient sound; The training process of the model for learning the wake-up speech features includes: Input the features of the wake-up speech signal, the features of the non-wake-up speech signal, and the identifier of the sleep state into a preset model; obtain a first estimated feature output by the preset model after performing wake-up speech enhancement on the features of the wake-up speech signal, and obtain a second estimated feature output by the preset model after performing wake-up speech enhancement on the features of the non-wake-up speech signal; generate a first loss function value according to the difference between the first estimated feature and the corresponding clean speech feature, and the difference between the second estimated feature and the corresponding clean speech feature; identifying the first estimated feature and outputting a first estimated state, and identifying the second estimated feature and outputting a second estimated state; generating a second loss function value according to whether the first estimated state is an awake state and whether the second estimated state is a sleep state; According to the first loss function value and the second loss function value, the model parameters of the preset model are updated to obtain a model that learns the wake-up speech features.

2. The method according to claim 1, characterized in that The method further comprises: When the current state is the awake state, performing overall speech enhancement on the speech features of the current ambient sound, wherein the overall speech enhancement represents enhancing the pure speech features in the speech features of the current ambient sound; According to the speech features of the current ambient sound after overall speech enhancement, a response signal of the current ambient sound is acquired and output.

3. The method according to claim 1, characterized in that: In a case where the current state is a dormant state, identifying whether the current ambient sound includes a wake-up voice includes: When the current state is a dormant state, obtaining a voice feature of the current ambient sound; According to the wake-up speech feature, the speech feature of the current ambient sound is enhanced for the wake-up speech; According to the voice feature of the current ambient sound enhanced by the wake-up voice, it is identified whether the current ambient sound includes the wake-up voice.

4. The method according to claim 3, characterized in that When the first-level recognition result indicates that the current ambient sound includes a wake-up voice, sending the current ambient sound to a server includes: In a case where the first-level recognition result indicates that the current ambient sound includes a wake-up speech, the speech features of the current ambient sound enhanced by the wake-up speech are sent to a server.

5. The method according to claim 3, characterized in that: The training process of the model for learning the wake-up speech features also includes: The feature of the wake-up voice signal and the clean voice feature contained therein are obtained, and the feature of the non-wake-up voice signal and the clean voice feature contained therein are obtained, wherein the clean voice feature of the wake-up voice signal includes the wake-up voice feature.

6. The method according to claim 5, characterized in that The training process of the model for learning the wake-up speech features further includes the following steps: In the process of updating the model parameters of the preset model, the characteristics of the sample speech signal and the pure speech characteristics contained therein, as well as the identification of the awakening state are input into the preset model after the model parameters are updated; Obtaining a third estimated feature outputted by the preset model after the model parameters are updated after overall speech enhancement is performed on the features of the sample speech signal; Generating a third loss function value according to the difference between the third estimated feature and the pure speech feature; According to the third loss function value, the preset model after the parameter update is updated to obtain a model that has learned the wake-up speech features, wherein the model that has learned the wake-up speech features is also used to perform overall speech enhancement on the speech features of the current ambient sound when the current state is the wake-up state, including: when the current state of the voice wake-up function is the wake-up state, the speech features of the current ambient sound are input into the model that has learned the wake-up speech features to perform overall speech enhancement.

7. The method according to claim 1, characterized in that Identifying whether the current ambient sound includes a wake-up voice includes: Identify, by a pre-trained first wake-up module, whether the voice feature of the current ambient sound includes a wake-up voice feature, wherein the first wake-up module is trained using the wake-up voice as a training sample; The second-level recognition result returned by the server is obtained by recognition through a pre-trained second wake-up module, where the second wake-up module is first trained with a general corpus as a training sample and then trained with a wake-up voice as a training sample; Among them, the probability that the first wake-up module recognizes that the voice feature of the current ambient sound includes the wake-up voice feature is higher than the probability that the second wake-up module recognizes that the current ambient sound meets the target wake-up condition.

8. A voice wake-up device, characterized in that: The device comprises: An acquisition module, configured to acquire the current state of the voice wake-up function and the current ambient sound; a recognition module configured to, when the current state is a dormant state, identify whether the current ambient sound includes a wake-up voice, and obtain a first-level recognition result; a sending module, configured to send the current ambient sound to a server when the first-level recognition result indicates that the current ambient sound includes a wake-up voice; A receiving module is configured to receive a second-level recognition result returned by the server, where the second-level recognition result is a result of the server identifying whether the current ambient sound meets a target wake-up condition; A state switching module is configured to switch the current state from the dormant state to the awake state when it is determined that the second-level recognition result indicates that the current ambient sound meets the target awakening condition; When the current state is a dormant state, the device is further configured to: input a speech feature of the current ambient sound into a model that has learned a wake-up speech feature, and obtain the current ambient sound after wake-up speech enhancement output by the model that has learned the wake-up speech feature, wherein the wake-up speech enhancement characterizes the enhancement of the wake-up speech feature in the speech feature of the current ambient sound; The training process of the model for learning the wake-up speech features includes: Input the features of the wake-up speech signal, the features of the non-wake-up speech signal, and the identifier of the sleep state into a preset model; obtain a first estimated feature output by the preset model after performing wake-up speech enhancement on the features of the wake-up speech signal, and obtain a second estimated feature output by the preset model after performing wake-up speech enhancement on the features of the non-wake-up speech signal; generate a first loss function value according to the difference between the first estimated feature and the corresponding clean speech feature, and the difference between the second estimated feature and the corresponding clean speech feature; identifying the first estimated feature and outputting a first estimated state, and identifying the second estimated feature and outputting a second estimated state; generating a second loss function value according to whether the first estimated state is an awake state and whether the second estimated state is a sleep state; According to the first loss function value and the second loss function value, the model parameters of the preset model are updated to obtain a model that learns the wake-up speech features.

9. The device according to claim 8, characterized in that The device also includes: An overall enhancement module is configured to perform overall speech enhancement on the speech features of the current ambient sound when the current state is the awake state, wherein the overall speech enhancement represents enhancement of pure speech features in the speech features of the current ambient sound; The response module is configured to acquire and output a response signal of the current ambient sound according to the voice features of the current ambient sound after overall voice enhancement.

10. The device according to claim 8, characterized in that The identification module comprises: an acquisition unit, configured to acquire the voice feature of the current ambient sound when the current state is a dormant state; a wake-up speech enhancement unit, configured to perform wake-up speech enhancement on the speech feature of the current ambient sound according to the wake-up speech feature; The recognition unit is configured to recognize whether the current ambient sound includes the wake-up speech according to the speech feature of the current ambient sound enhanced by the wake-up speech.

11. The device according to claim 10, characterized in that The sending module comprises: The sending unit is configured to send the voice features of the current ambient sound enhanced by the wake-up speech to the server when the first-level recognition result indicates that the current ambient sound includes the wake-up speech.

12. The device according to claim 10, characterized in that The training process of the model for learning the wake-up speech features also includes: The feature of the wake-up voice signal and the clean voice feature contained therein are obtained, and the feature of the non-wake-up voice signal and the clean voice feature contained therein are obtained, wherein the clean voice feature of the wake-up voice signal includes the wake-up voice feature.

13. The device according to claim 12, characterized in that The training process of the model for learning the wake-up speech features further includes the following steps: In the process of updating the model parameters of the preset model, the characteristics of the sample speech signal and the pure speech characteristics contained therein, as well as the identification of the awakening state are input into the preset model after the model parameters are updated; Obtaining a third estimated feature outputted by the preset model after the model parameters are updated after overall speech enhancement is performed on the features of the sample speech signal; Generating a third loss function value according to the difference between the third estimated feature and the pure speech feature; According to the third loss function value, the preset model after the parameter update is updated to obtain a model that has learned the wake-up speech features, wherein the model that has learned the wake-up speech features is also used to perform overall speech enhancement on the speech features of the current ambient sound when the current state is the wake-up state, including: when the current state of the voice wake-up function is the wake-up state, the speech features of the current ambient sound are input into the model that has learned the wake-up speech features to perform overall speech enhancement.

14. The device according to claim 8, characterized in that The identification module comprises: A first recognition unit is configured to recognize whether the voice feature of the current ambient sound includes a wake-up voice feature through a pre-trained first wake-up module, wherein the first wake-up module is trained using the wake-up voice as a training sample; The second-level recognition result returned by the server and received by the receiving module is obtained by recognition by a pre-trained second wake-up module, wherein the second wake-up module is first trained with a general corpus as a training sample and then trained with a wake-up speech as a training sample; Among them, the probability that the first wake-up module recognizes that the voice feature of the current ambient sound includes the wake-up voice feature is higher than the probability that the second wake-up module recognizes that the current ambient sound meets the target wake-up condition.

15. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the voice wake-up method as described in any one of claims 1 to 7. 16 . A computer-readable storage medium, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the voice wake-up method according to any one of claims 1 to 7.

17. A computer program product comprising a computer program or computer instructions, characterized in that When the computer program or computer instruction is executed by a processor, the voice wake-up method according to any one of claims 1 to 7 is implemented.

Images