Voice wake-up method, device, equipment and storage medium

Abstract

The present invention provides a voice wake-up method, device, equipment and storage medium, the method comprising: identifying historical sampling data collected by a sound pickup device, obtaining recognition result information for characterizing whether the historical sampling data includes voice; determining a target voice wake-up threshold corresponding to the recognition result information based on a preset voice wake-up threshold; and performing wake-up processing based on the target voice wake-up threshold. The voice wake-up method, device, equipment and storage medium provided by the present invention can improve the reliability of voice wake-up.

Description

Technical Field

[0001] The present invention relates to the field of voice processing technology, and in particular to a voice wake-up method, device, equipment and storage medium. Background Art

[0002] With the development of voice technology, many electronic devices have introduced voice recognition technology, so that they can be awakened by voice to control the electronic devices to switch from a non-working state to a working state and perform work.

[0003] At present, in order to improve the success rate of electronic device wake-up and reduce the false wake-up rate of electronic devices, a common practice is to identify the wake-up words in voice information through a wake-up model. When the confidence of the identified wake-up words meets the preset voice wake-up threshold, the electronic device wakes up successfully.

[0004] However, in the above wake-up methods, false wake-ups often occur or the electronic device cannot be woken up even if the wake-up word is mentioned. Therefore, how to improve the reliability of voice wake-up is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The present invention provides a voice wake-up method, device, equipment and storage medium, which are used to solve the defect of low reliability of voice wake-up in the prior art and improve the reliability of voice wake-up.

[0006] The present invention provides a voice wake-up method, comprising:

[0007] Identify the historical sampling data collected by the sound pickup device to obtain recognition result information indicating whether the historical sampling data includes speech;

[0008] Based on a preset voice wake-up threshold, determining a target voice wake-up threshold corresponding to the recognition result information;

[0009] Based on the target voice wake-up threshold, wake-up processing is performed.

[0010] According to a voice wake-up method provided by the present invention, determining a target voice wake-up threshold corresponding to the recognition result information based on a preset voice wake-up threshold includes:

[0011] In a case where the recognition result information indicates that the historical sampling data includes voice, determining the preset voice wake-up threshold as the target voice wake-up threshold, or determining the preset voice wake-up threshold that has been adjusted higher as the target voice wake-up threshold;

[0012] In a case where the recognition result information indicates that the historical sampling data does not include voice, the preset voice wake-up threshold that has been adjusted down is determined as the target voice wake-up threshold.

[0013] According to a voice wake-up method provided by the present invention, the recognition result information includes a voice confidence level for characterizing that voice is included in the historical sampling data, and a non-voice confidence level for characterizing that non-voice is included in the historical sampling data;

[0014] In the case where the recognition result information indicates that the historical sampling data includes voice, determining the preset voice wake-up threshold as the target voice wake-up threshold, or determining the preset voice wake-up threshold that has been adjusted higher as the target voice wake-up threshold, includes:

[0015] In a case where the voice confidence is greater than or equal to the first confidence threshold and less than the second confidence threshold, determining the preset voice wake-up threshold that has been adjusted up as the target voice wake-up threshold;

[0016] In a case where the speech confidence is greater than or equal to the second confidence threshold, and the non-speech confidence is less than a third confidence threshold, determining the preset speech wake-up threshold as the target speech wake-up threshold;

[0017] The first confidence threshold is smaller than the second confidence threshold, and the third confidence threshold is smaller than the second confidence threshold.

[0018] According to a voice wake-up method provided by the present invention, when the voice confidence is less than the first confidence threshold, the recognition result information indicates that the historical sampling data does not include voice; and

[0019] The recognition result information also includes a silence confidence level for characterizing that the historical sampling data contains neither speech nor non-speech. When the silence confidence level is greater than or equal to a fourth confidence threshold, the recognition result information characterizes that the historical sampling data does not contain speech.

[0020] According to a voice wake-up method provided by the present invention, the identifying of historical sampling data collected by a microphone to obtain recognition result information for characterizing whether the historical sampling data includes voice includes:

[0021] Inputting the historical sampling data into a speech recognition model to obtain the recognition result information;

[0022] The speech recognition model is obtained by training an initial speech recognition model based on multiple audio samples, wherein the multiple audio samples include audio samples containing speech, audio samples containing non-speech, and audio samples containing neither speech nor non-speech.

[0023] According to a voice wake-up method provided by the present invention,

[0024] The identifying of the historical sampling data collected by the sound pickup device includes:

[0025] The historical sampling data collected by the sound pickup device within a preset time period with the current time as the end time is identified.

[0026] According to a voice wake-up method provided by the present invention, the wake-up process is performed based on the target voice wake-up threshold, including:

[0027] Acquire current sampling data collected by the sound pickup device at the current moment;

[0028] Obtaining the wake-up confidence that the current sampled data includes a preset wake-up word, and the current sound intensity corresponding to the current sampled data;

[0029] When the wake-up confidence is greater than or equal to the target voice wake-up threshold and the current sound intensity is greater than the target sound intensity threshold, a wake-up operation is performed.

[0030] According to a voice wake-up method provided by the present invention, the method further includes:

[0031] Obtaining historical sound intensity corresponding to the historical sampling data;

[0032] Based on the historical sound intensity, the target sound intensity threshold is determined.

[0033] The present invention also provides a voice wake-up device, comprising:

[0034] A recognition module, used to recognize historical sampling data collected by a microphone, and obtain recognition result information for indicating whether the historical sampling data includes speech;

[0035] A determination module, configured to determine a target voice wake-up threshold corresponding to the recognition result information based on a preset voice wake-up threshold;

[0036] A processing module is used to perform wake-up processing based on the target voice wake-up threshold.

[0037] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-mentioned voice wake-up methods is implemented.

[0038] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the voice wake-up method described in any one of the above is implemented.

[0039] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the voice wake-up method described above is implemented.

[0040] The voice wake-up method, device, equipment and storage medium provided by the embodiments of the present invention identify the historical sampling data collected by the sound pickup device, obtain the recognition result information used to characterize whether the historical sampling data includes voice, and then determine the target voice wake-up threshold corresponding to the recognition result information based on the preset voice wake-up threshold, and perform wake-up processing based on the target voice wake-up threshold. Since different target voice wake-up thresholds are set in scenes including voice and scenes not including voice, the corresponding target voice wake-up threshold can be determined according to the recognition result of whether the historical sampling data contains voice to perform wake-up processing, thereby reducing the false wake-up rate and improving the wake-up success rate, thereby improving the reliability of voice wake-up. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 A schematic diagram of an application scenario of the voice wake-up method provided by an embodiment of the present invention;

[0043] Figure 2 One of the flowcharts of the voice wake-up method provided by an embodiment of the present invention;

[0044] Figure 3 A second flowchart of the voice wake-up method provided by an embodiment of the present invention;

[0045] Figure 4 A schematic diagram of a voice wake-up device provided by an embodiment of the present invention;

[0046] Figure 5 It is a schematic diagram of the physical structure of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Voice wake-up technology is an important branch of voice recognition technology. Voice wake-up technology monitors the user's voice to determine whether the user has said the specified wake-up word, and then performs wake-up processing. It is currently used in vehicles, navigation, smart homes, etc. In actual use, voice information may contain some words similar to the wake-up word. When the user mentions these similar words during the chat, the electronic device may be woken up by mistake. Or in a noisy environment, even if the user says the wake-up word, the electronic device may not be woken up. Therefore, there is a high false wake-up rate or a low wake-up success rate, resulting in poor reliability of voice wake-up.

[0049] Taking the above problems into consideration, an embodiment of the present invention provides a voice wake-up method. The technical concept of the method is that in different scenarios where the current environment contains voice or does not contain voice, the recognition accuracy of the electronic device for the wake-up word may be different. Therefore, different target voice wake-up thresholds can be determined based on whether the recognition result information of voice is included in the historical sampling data to match different scenarios. In this way, in the process of wake-up processing based on the target voice wake-up threshold, the false wake-up rate can be reduced and the wake-up success rate can be increased, thereby improving the reliability of voice wake-up.

[0050] Combine the following Figure 1-Figure 3 The technical solution of the voice wake-up method of the present invention is described.

[0051] For example, Figure 1 A schematic diagram of an application scenario of the voice wake-up method provided by an embodiment of the present invention, such as Figure 1As shown, assuming that the wake-up word of the electronic device 103 is "Xiao Ling Xiao Ling", when the user says the above wake-up word, the electronic device 103 can be started. In actual applications, during the chat between user 101 and user 102, it is assumed that "it's just a little win, a little win". If the electronic device 103 is in a relatively noisy environment, or the user 101 is far away from the electronic device 103, the electronic device 103 is likely to recognize "Xiao Ying Xiao Ying" as "Xiao Ling Xiao Ling" and wake up. In order to avoid this phenomenon, in the present invention, when the electronic device 103 recognizes that the historical sampling data contains voice, that is, when the user 101 and the user 102 are chatting, the target voice wake-up threshold in the scene can be determined based on the preset voice wake-up threshold, for example, the preset voice wake-up threshold can be increased, or the preset voice wake-up threshold can be maintained unchanged, so that the electronic device 103 can be prevented from being woken up by mistake. When the electronic device 103 recognizes that the historical sampling data does not contain voice, that is, after the chat between the user 101 and the user 102 ends, the target voice wake-up threshold in the scenario of no voice can be re-determined, for example, the preset voice wake-up threshold can be lowered, so as to ensure the wake-up success rate of the electronic device 103. By dynamically adjusting the target voice wake-up threshold, the reliability of voice wake-up can be improved.

[0052] The voice wake-up method provided by the embodiment of the present invention can be applied to Figure 1 In the electronic device 103 shown, Figure 1 The electronic device in the description is only an example. In actual applications, the electronic device can be the device itself that is awakened, or other devices other than the device that perform wake-up recognition and control the awakening of the device. The electronic device can be any device with voice control function, such as a mobile terminal, a smart speaker, a smart air conditioner, etc., and the other device that performs wake-up recognition and controls the awakening of the electronic device can be a server, a mobile terminal, etc.

[0053] Figure 2 One of the flow charts of the voice wake-up method provided by the embodiment of the present invention is as follows: Figure 2 As shown, the method includes:

[0054] Step 201: Identify the historical sampling data collected by the sound pickup device to obtain recognition result information for indicating whether the historical sampling data includes speech.

[0055] The sound pickup device provided in the electronic device may be, for example, a microphone or a microphone array, or other devices capable of collecting sound information.

[0056] In this step, when the historical sampling data collected by the sound pickup device is identified, the historical sampling data can be framed to determine the recognition result information corresponding to each frame of data. The recognition result information corresponding to each frame of data is summarized to obtain the recognition result information of the historical sampling data, wherein the recognition result information is used to indicate whether the historical sampling data includes speech.

[0057] Step 202: Based on a preset voice wake-up threshold, determine a target voice wake-up threshold corresponding to the recognition result information.

[0058] Among them, the preset voice wake-up threshold can be understood as the initial voice wake-up threshold, or can also be understood as the voice wake-up threshold set by the electronic device when it leaves the factory.

[0059] In this step, since the electronic device may have different recognition accuracy for wake-up words in different scenarios. For example, in scenarios where voice exists, such as multi-person chats or multi-person meetings, the electronic device may recognize non-wake-up words as wake-up words, resulting in false wake-up, while in scenarios where there is silence or no voice, the probability of false wake-up is relatively low. Therefore, the electronic device can set different target voice wake-up thresholds based on preset voice wake-up thresholds for different scenarios. For example, in a scenario containing voice, the target voice wake-up threshold is determined to be threshold 1, and in a scenario not containing voice, the target voice wake-up threshold is determined to be threshold 2, where threshold 1 is greater than threshold 2.

[0060] In addition, in order to avoid adjusting the target voice wake-up threshold back and forth, the electronic device can determine the recognition result information of the historical sampling data based on the historical sampling data, and then determine the target voice wake-up threshold corresponding to the recognition result information based on the preset voice wake-up threshold when it is determined that the recognition result information maintains the preset time without change. The preset time can be set based on experience, for example, it can be set to 5s.

[0061] Step 203: Perform wake-up processing based on the target voice wake-up threshold.

[0062] In this step, after determining the target voice wake-up threshold, it is determined whether to wake up the electronic device based on the target voice wake-up threshold. Exemplarily, the wake-up confidence of the preset wake-up word contained in the current sampling data collected at the current moment can be determined by the voice wake-up model, and it is determined whether the wake-up confidence is higher than the target voice wake-up threshold. If the wake-up confidence is higher than the target voice wake-up threshold, the electronic device is woken up, otherwise, the electronic device is not woken up.

[0063] It should be understood that since the above-mentioned target voice wake-up threshold is dynamically updated, before a new target voice wake-up threshold is determined, the electronic device will perform wake-up processing based on the target voice wake-up threshold.

[0064] The voice wake-up method provided in the embodiment of the present invention identifies the historical sampling data collected by the sound pickup device, obtains the recognition result information used to characterize whether the historical sampling data includes voice, and then determines the target voice wake-up threshold corresponding to the recognition result information based on the preset voice wake-up threshold, and performs wake-up processing based on the target voice wake-up threshold. Since different target voice wake-up thresholds are set in the scenes including voice and excluding voice, respectively, the corresponding target voice wake-up threshold can be determined according to the recognition result of whether the historical sampling data contains voice to perform wake-up processing, thereby reducing the false wake-up rate and improving the wake-up success rate, thereby improving the reliability of voice wake-up. In addition, by dynamically adjusting the preset voice wake-up threshold, the false wake-up rate of voice wake-up can be reduced without changing the accuracy of the wake-up model.

[0065] Figure 3 The second flow chart of the voice wake-up method provided by the embodiment of the present invention is as follows. Figure 2 Based on the embodiment shown, the process of determining the target voice wake-up threshold corresponding to the recognition result information based on the preset voice wake-up threshold in step 202 is described in detail. Figure 3 As shown, the method includes:

[0066] Step 301: Identify the historical sampling data collected by the sound pickup device to obtain recognition result information indicating whether the historical sampling data includes speech.

[0067] In one possible implementation, when historical sampling data is recognized, the historical sampling data can be input into a speech recognition model to obtain the recognition result information; wherein the speech recognition model is obtained by training an initial speech recognition model based on multiple audio samples, and the multiple audio samples include audio samples containing speech, audio samples containing non-speech, and audio samples containing neither speech nor non-speech.

[0068] Specifically, the historical sampling data can be input into a pre-trained speech recognition model to output recognition result information. The speech recognition model can be a deep neural network (DNN).

[0069] Among them, the above-mentioned recognition result information may include a speech confidence used to characterize that speech is included in the historical sampling data, and a non-speech confidence used to characterize that non-speech is included in the historical sampling data; or, the recognition result information may include a speech confidence used to characterize that speech is included in the historical sampling data, a non-speech confidence used to characterize that non-speech is included in the historical sampling data, and a silence confidence used to characterize that neither speech nor non-speech is included in the historical sampling data.

[0070] The above speech recognition model can be trained in the following way:

[0071] First, a large number of audio samples are collected, including audio samples containing speech, audio samples containing non-speech, and audio samples containing neither speech nor non-speech. It should be understood that in order to enable the speech recognition model to recognize historical sampling data in various scenarios, the above audio samples can be collected in various scenarios, such as chat scenarios, meeting discussion scenarios, and outdoor scenarios. After the audio samples are collected, each audio sample can be labeled, such as containing speech, containing non-speech, or silence.

[0072] Input the collected audio samples into the initial speech recognition model to obtain the prediction results corresponding to each audio sample, and then compare the prediction results with the annotation information of the audio sample to obtain the loss information. Adjust the model parameters of the initial speech recognition model based on the loss information, and repeat the above process until the obtained speech recognition model converges or the loss information is minimized, and the final model is determined as the speech recognition model.

[0073] In this embodiment, the historical sampling data can be input into the speech recognition model to obtain the recognition result information, thereby improving the efficiency of determining the recognition result information.

[0074] Step 302: When the recognition result information indicates that the historical sampling data includes speech, a preset speech wake-up threshold is determined as a target speech wake-up threshold, or a preset speech wake-up threshold that has been adjusted higher is determined as a target speech wake-up threshold.

[0075] Optionally, the recognition result information includes a speech confidence for characterizing that speech is included in the historical sampling data, and a non-speech confidence for characterizing that non-speech is included in the historical sampling data. Therefore, whether speech is included in the historical sampling data can be determined based on the size of the speech confidence and the non-speech confidence. For example, when the speech confidence is greater than or equal to a first confidence threshold, it can be determined that the historical sampling data includes speech, and when the speech confidence is less than the first confidence threshold, it can be determined that the historical sampling data does not include speech.

[0076] It should be understood that the above-mentioned speech confidence can be understood as a score for the speech in the historical sampling data, which is positively correlated with the speech clarity and / or the sound intensity of the speech. For example, if the historical sampling data contains near-field speech, or the contained speech is relatively clear, or the sound intensity is relatively high, then the speech confidence in the historical sampling data is high, otherwise, the speech confidence in the historical sampling data is low.

[0077] Furthermore, in the case where the recognition result information represents that the historical sampling data includes voice, the voice may be near-field voice or far-field voice, or the voice may be relatively clear or relatively vague, and these contents can be represented by the voice confidence in the recognition result information. Therefore, for the above different situations, the electronic device can determine the target voice wake-up threshold in different ways based on the size of the voice confidence.

[0078] Exemplarily, when the speech confidence is greater than or equal to the first confidence threshold and less than the second confidence threshold, the preset speech wake-up threshold that has been adjusted up is determined as the target speech wake-up threshold; when the speech confidence is greater than or equal to the second confidence threshold and the non-speech confidence is less than the third confidence threshold, the preset speech wake-up threshold is determined as the target speech wake-up threshold; wherein the first confidence threshold is less than the second confidence threshold, and the third confidence threshold is less than the second confidence threshold.

[0079] Specifically, for the convenience of description, the scenario in which the speech confidence is greater than or equal to the first confidence threshold and less than the second confidence threshold is referred to as the first scenario, and the scenario in which the speech confidence is greater than or equal to the second confidence threshold and the non-speech confidence is less than the third confidence threshold is referred to as the second scenario. In the first scenario, since the speech confidence is greater than or equal to the first confidence threshold and less than the second confidence threshold, it means that the speech contained in the historical sampling data is relatively vague, or the sound intensity is not high. Therefore, in this first scenario, the electronic device may recognize a non-wake-up word as a wake-up word, resulting in false wake-up. In order to solve this problem, in an embodiment of the present invention, the preset speech wake-up threshold can be increased in the first scenario, and the preset speech wake-up threshold that has been increased can be determined as the target speech wake-up threshold.

[0080] In actual applications, the preset voice wake-up threshold can be increased by a preset value, such as increasing the preset voice wake-up threshold by 0.1, or the preset voice wake-up threshold can be increased according to the sound intensity of the voice contained in the historical sampling data. For example, the greater the sound intensity, the greater the increase.

[0081] In addition, in the second scenario, since the speech confidence is greater than or equal to the second confidence threshold, and the non-speech confidence is less than the third confidence threshold, it means that the speech contained in the historical sampling data is relatively clear or has a higher sound intensity, while the non-speech contained is relatively vague or has a lower sound intensity. Therefore, the probability of false awakening rate in the second scenario is not high. In order to improve the wake-up success rate of the electronic device and save the power consumption of the electronic device, the preset speech awakening threshold will not be processed at this time, that is, the preset speech awakening threshold is directly determined as the target speech awakening threshold.

[0082] Of course, in order to further reduce the false wake-up rate, in the specific implementation process, in the second scenario, the preset voice wake-up threshold can also be increased, but the adjustment range of the preset voice wake-up threshold is usually smaller than the adjustment range of the preset voice wake-up threshold in the first scenario.

[0083] Among them, the specific values ​​of the first confidence threshold, the second confidence threshold and the third confidence threshold can be set according to actual conditions or experience. For example, the first confidence threshold can be set to 0.5, the second confidence threshold can be set to 0.8, and the third confidence threshold can be set to 0.4. The embodiment of the present invention does not limit the specific values ​​of each confidence threshold.

[0084] In this embodiment, when the voice confidence is greater than or equal to the first confidence threshold and less than the second confidence threshold, the preset voice wake-up threshold that has been adjusted up is determined as the target voice wake-up threshold, thereby reducing the false wake-up situation and reducing the false wake-up rate of the electronic device. When the voice confidence is greater than or equal to the second confidence threshold and the non-voice confidence is less than the third confidence threshold, the preset voice wake-up threshold is determined as the target voice wake-up threshold, which can not only improve the wake-up success rate of the electronic device, but also reduce the power consumption of the electronic device because the electronic device does not adjust the preset voice wake-up threshold.

[0085] Step 303: When the recognition result information indicates that the historical sampling data does not include speech, the preset speech wake-up threshold that has been adjusted down is determined as the target speech wake-up threshold.

[0086] Specifically, the scenario in which the historical sampling data does not include voice can be referred to as the third scenario, wherein the third scenario can include a scenario containing only non-voice and a silent scenario. It should be understood that in either of the above scenarios, since the historical sampling data does not include voice, the probability that the electronic device recognizes a non-wake-up word as a wake-up word is very small, that is, the false wake-up rate in the third scenario is low. In order to improve the wake-up success rate of the electronic device, in the third scenario, the preset voice wake-up threshold can be lowered, that is, the preset voice wake-up threshold that has been lowered is determined as the target voice wake-up threshold.

[0087] In actual applications, the preset voice wake-up threshold can be lowered by a preset value, such as lowering the preset voice wake-up threshold by 0.1. In the case of only non-voice, the preset voice wake-up threshold can also be lowered according to the sound intensity of the non-voice contained in the historical sampling data, such as the greater the sound intensity, the greater the reduction.

[0088] Wherein, in the case where the historical sampled data does not contain speech, the speech confidence may also be used to determine. Optionally, in the case where the speech confidence is less than the first confidence threshold, the recognition result information indicates that the historical sampled data does not contain speech.

[0089] In addition, the recognition result information also includes a silence confidence level used to characterize that the historical sampling data contains neither speech nor non-speech. When the silence confidence level is greater than or equal to the fourth confidence threshold, the recognition result information characterizes that the historical sampling data does not include speech.

[0090] It should be understood that when the recognition result information includes speech confidence, non-speech confidence and silence confidence, the sum of the three should be 1. Therefore, when the silence confidence is greater than or equal to the fourth confidence threshold, it means that both the speech confidence and the non-speech confidence are low. At this time, it can be determined that the scene is silent. The specific values ​​of the fourth confidence threshold and the third confidence threshold can be the same or different.

[0091] In order to improve the accuracy of silent scene determination, in a specific implementation process, it can also be determined that the current silent scene is in the case that the silent confidence is greater than or equal to the fourth confidence threshold and the speech confidence and non-speech confidence are both less than the fifth confidence threshold.

[0092] In this embodiment, when the speech confidence is less than the first confidence threshold, or when the silence confidence is greater than or equal to the fourth confidence threshold, it can be determined that the historical sampling data does not include speech, so that the scene classification method of whether the historical sampling data contains speech is simpler.

[0093] It should be noted that the first scene, the second scene and the third scene mentioned in the above steps 302 and 303 can also be determined by a pre-trained scene classification model. Specifically, the speech confidence, the non-speech confidence and the silence confidence can be input into the scene classification model, and the confidence corresponding to the first scene, the confidence corresponding to the second scene and the confidence corresponding to the third scene can be obtained, and the scene with the highest confidence among these three confidences is determined as the final scene. Among them, the scene classification model can be a DNN.

[0094] The scene classification model can be trained in the following manner: the speech confidence, non-speech confidence and silence confidence output by the speech recognition model are used as training samples, and each training sample is annotated, and the annotation information includes the result of scene classification. The training samples are input into the initial scene classification model to obtain the scene prediction result corresponding to each training sample, and the scene prediction result is compared with the annotation information to obtain loss information. Based on the loss information, the model parameters of the initial scene classification model are adjusted, and the above process is repeated until the obtained model converges or the loss information is minimized, and the final model is determined as the scene classification model.

[0095] Step 304: Perform wake-up processing based on the target voice wake-up threshold.

[0096] The voice wake-up method provided by an embodiment of the present invention determines a preset voice wake-up threshold as a target voice wake-up threshold, or determines a preset voice wake-up threshold that has been adjusted up as a target voice wake-up threshold, so that when wake-up processing is performed based on the target voice wake-up threshold, the phenomenon of mistakenly waking up the electronic device by words similar to the wake-up word can be avoided, thereby reducing the false wake-up rate of the electronic device. In addition, when the recognition result information indicates that the historical sampling data does not include voice, the preset voice wake-up threshold that has been adjusted down is determined as the target voice wake-up threshold, so that when wake-up processing is performed based on the target voice wake-up threshold, the phenomenon of failing to wake up the electronic device after speaking the wake-up word can be avoided, thereby improving the wake-up rate of the electronic device.

[0097] On the basis of any of the above embodiments, in order to ensure the validity of the target voice wake-up threshold, when identifying the historical sampling data collected by the sound pickup device, the historical sampling data collected by the sound pickup device within a preset time period with the current moment as the end moment may be identified.

[0098] The preset time period can be set based on experience, for example, it can be set to 5s or 8s, etc. The smaller the preset time period is set, the higher the reliability of voice wake-up of the electronic device.

[0099] On the basis of the above embodiment, when the wake-up processing is performed based on the target voice wake-up threshold, the current sampling data collected by the sound pickup device at the current moment can be obtained, and the wake-up confidence of the current sampling data containing the preset wake-up word and the current sound intensity corresponding to the current sampling data can be obtained; when the wake-up confidence is greater than or equal to the target voice wake-up threshold and the current sound intensity is greater than the target sound intensity threshold, the wake-up operation is performed.

[0100] Specifically, the wake-up confidence can be used to characterize the probability that the current sampled data contains a preset wake-up word. The higher the wake-up confidence, the greater the probability that the current sampled data contains a preset wake-up word. When the electronic device determines that the wake-up confidence is greater than or equal to the target voice wake-up threshold, it indicates that the current sampled data may contain a preset wake-up word.

[0101] Furthermore, when users use preset wake-up words to wake up electronic devices, they usually say the preset wake-up words at a relatively high volume. Therefore, when the wake-up confidence is greater than or equal to the target voice wake-up threshold, it is also possible to determine whether the current sound intensity is greater than the target sound intensity threshold. If it is greater, it means that the user has said the preset wake-up word, and the electronic device will perform the wake-up operation. Otherwise, the electronic device will not be woken up.

[0102] In the above method, when the wake-up confidence of the preset wake-up word contained in the current sampling data is greater than or equal to the target voice wake-up threshold, and the current sound intensity corresponding to the current sampling data is greater than the target sound intensity threshold, the wake-up operation is performed. Since whether the preset wake-up word is contained in the current sampling data is judged from different dimensions, the false wake-up rate of the electronic device can be further reduced.

[0103] The target sound intensity threshold may be calculated in real time, and may be determined, for example, by obtaining historical sound intensities corresponding to historical sampling data and based on the historical sound intensities.

[0104] For example, the average value of the historical sound intensities corresponding to the historical sampling data within a preset time period may be determined as the target sound intensity threshold, or the maximum value of the historical sound intensities corresponding to the historical sampling data may be determined as the target sound intensity threshold.

[0105] In the above embodiment, the target sound intensity threshold is determined by the historical sound intensity corresponding to the historical sampling data, so that the target sound intensity threshold can be dynamically adjusted based on the sound intensity in the environment, thereby further reducing the false awakening rate of the electronic device.

[0106] The voice wake-up device provided by the present invention is described below. The voice wake-up device described below and the voice wake-up method described above can be referenced to each other.

[0107] Figure 4 A schematic diagram of a voice wake-up device provided by an embodiment of the present invention, such as Figure 4 As shown, the device comprises:

[0108] The recognition module 11 is used to recognize the historical sampling data collected by the microphone and obtain recognition result information for indicating whether the historical sampling data includes speech;

[0109] A determination module 12, configured to determine a target voice wake-up threshold corresponding to the recognition result information based on a preset voice wake-up threshold;

[0110] The processing module 13 is used to perform wake-up processing based on the target voice wake-up threshold.

[0111] The device of this embodiment can be used to execute the method of any of the embodiments in the aforementioned electronic device side method embodiments. Its specific implementation process and technical effects are similar to those in the electronic device side method embodiments. For details, please refer to the detailed introduction in the electronic device side method embodiments, which will not be repeated here.

[0112] Optionally, the determining module 12 is specifically configured to:

[0113] In a case where the recognition result information indicates that the historical sampling data includes voice, determining the preset voice wake-up threshold as the target voice wake-up threshold, or determining the preset voice wake-up threshold that has been adjusted higher as the target voice wake-up threshold;

[0114] In a case where the recognition result information indicates that the historical sampling data does not include voice, the preset voice wake-up threshold that has been adjusted down is determined as the target voice wake-up threshold.

[0115] Optionally, the recognition result information includes a voice confidence level for characterizing that voice is contained in the historical sample data, and a non-voice confidence level for characterizing that non-voice is contained in the historical sample data;

[0116] The determining module 12 is specifically used for:

[0117] In a case where the voice confidence is greater than or equal to the first confidence threshold and less than the second confidence threshold, determining the preset voice wake-up threshold that has been adjusted up as the target voice wake-up threshold;

[0118] In a case where the speech confidence is greater than or equal to the second confidence threshold, and the non-speech confidence is less than a third confidence threshold, determining the preset speech wake-up threshold as the target speech wake-up threshold;

[0119] The first confidence threshold is smaller than the second confidence threshold, and the third confidence threshold is smaller than the second confidence threshold.

[0120] Optionally, when the speech confidence is less than the first confidence threshold, the recognition result information indicates that the historical sampling data does not include speech; and

[0121] The recognition result information also includes a silence confidence level for characterizing that the historical sampling data contains neither speech nor non-speech. When the silence confidence level is greater than or equal to a fourth confidence threshold, the recognition result information characterizes that the historical sampling data does not contain speech.

[0122] Optionally, the identification module 11 is specifically used to:

[0123] Inputting the historical sampling data into a speech recognition model to obtain the recognition result information;

[0124] The speech recognition model is obtained by training an initial speech recognition model based on multiple audio samples, wherein the multiple audio samples include audio samples containing speech, audio samples containing non-speech, and audio samples containing neither speech nor non-speech.

[0125] Optionally, the identification module 11 is specifically used to:

[0126] The historical sampling data collected by the sound pickup device within a preset time period with the current time as the end time is identified.

[0127] Optionally, the processing module 13 is specifically configured to:

[0128] Acquire current sampling data collected by the sound pickup device at the current moment;

[0129] Obtaining the wake-up confidence that the current sampled data includes a preset wake-up word, and the current sound intensity corresponding to the current sampled data;

[0130] When the wake-up confidence is greater than or equal to the target voice wake-up threshold and the current sound intensity is greater than the target sound intensity threshold, a wake-up operation is performed.

[0131] Optionally, the device further includes an acquisition module; wherein:

[0132] An acquisition module, used to acquire the historical sound intensity corresponding to the historical sampling data;

[0133] The determination module 12 is configured to determine the target sound intensity threshold based on the historical sound intensity.

[0134] The device of this embodiment can be used to execute the method of any of the embodiments in the aforementioned electronic device side method embodiments. Its specific implementation process and technical effects are similar to those in the electronic device side method embodiments. For details, please refer to the detailed introduction in the electronic device side method embodiments, which will not be repeated here.

[0135] Figure 5is a schematic diagram of the physical structure of an electronic device provided by the present invention, such as Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the voice wake-up method, which includes: identifying the historical sampling data collected by the sound pickup device, obtaining the recognition result information for characterizing whether the historical sampling data includes voice; based on the preset voice wake-up threshold, determining the target voice wake-up threshold corresponding to the recognition result information; and performing the wake-up process based on the target voice wake-up threshold.

[0136] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0137] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the voice wake-up method provided by the above methods. The method includes: identifying historical sampling data collected by a sound pickup device to obtain recognition result information for characterizing whether the historical sampling data includes voice; based on a preset voice wake-up threshold, determining a target voice wake-up threshold corresponding to the recognition result information; and performing wake-up processing based on the target voice wake-up threshold.

[0138] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the voice wake-up method provided by the above-mentioned methods. The method includes: identifying historical sampling data collected by a sound pickup device to obtain recognition result information for characterizing whether the historical sampling data includes voice; based on a preset voice wake-up threshold, determining a target voice wake-up threshold corresponding to the recognition result information; and performing wake-up processing based on the target voice wake-up threshold.

[0139] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0140] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A voice wake-up method, characterized in that: include: Identify the historical sampling data collected by the sound pickup device to obtain recognition result information indicating whether the historical sampling data includes speech; Based on a preset voice wake-up threshold, determining a target voice wake-up threshold corresponding to the recognition result information; Performing wake-up processing based on the target voice wake-up threshold; The determining, based on a preset voice wake-up threshold, a target voice wake-up threshold corresponding to the recognition result information includes: In a case where the recognition result information indicates that the historical sampling data includes voice, determining the preset voice wake-up threshold as the target voice wake-up threshold, or determining the preset voice wake-up threshold that has been adjusted higher as the target voice wake-up threshold; In a case where the recognition result information indicates that the historical sampling data does not include voice, the preset voice wake-up threshold that has been adjusted down is determined as the target voice wake-up threshold.

2. The voice wake-up method according to claim 1, characterized in that: The recognition result information includes a voice confidence level for characterizing that voice is included in the historical sample data, and a non-voice confidence level for characterizing that non-voice is included in the historical sample data; In the case where the recognition result information indicates that the historical sampling data includes voice, determining the preset voice wake-up threshold as the target voice wake-up threshold, or determining the preset voice wake-up threshold that has been adjusted higher as the target voice wake-up threshold, includes: When the voice confidence is greater than or equal to the first confidence threshold and less than the second confidence threshold, the preset voice wake-up threshold that has been adjusted up is determined as the target voice wake-up threshold; When the speech confidence is greater than or equal to the second confidence threshold, and the non-speech confidence is less than a third confidence threshold, determining the preset speech wake-up threshold as the target speech wake-up threshold; The first confidence threshold is smaller than the second confidence threshold, and the third confidence threshold is smaller than the second confidence threshold.

3. The voice wake-up method according to claim 2, characterized in that: In a case where the speech confidence is less than the first confidence threshold, the recognition result information indicates that the historical sampling data does not include speech; and The recognition result information also includes a silence confidence level for characterizing that the historical sampling data contains neither speech nor non-speech. When the silence confidence level is greater than or equal to a fourth confidence threshold, the recognition result information characterizes that the historical sampling data does not contain speech.

4. The voice wake-up method according to claim 2 or 3, characterized in that: Identifying the historical sampling data collected by the microphone to obtain recognition result information for indicating whether the historical sampling data includes speech includes: Inputting the historical sampling data into a speech recognition model to obtain the recognition result information; The speech recognition model is obtained by training an initial speech recognition model based on multiple audio samples, wherein the multiple audio samples include audio samples containing speech, audio samples containing non-speech, and audio samples containing neither speech nor non-speech.

5. The voice wake-up method according to any one of claims 1 to 3, characterized in that: The identifying of the historical sampling data collected by the sound pickup device includes: The historical sampling data collected by the sound pickup device within a preset time period with the current time as the end time is identified.

6. The voice wake-up method according to claim 5, characterized in that: The performing wake-up processing based on the target voice wake-up threshold includes: Acquire current sampling data collected by the sound pickup device at the current moment; Obtaining the wake-up confidence that the current sampled data includes a preset wake-up word, and the current sound intensity corresponding to the current sampled data; When the wake-up confidence is greater than or equal to the target voice wake-up threshold and the current sound intensity is greater than the target sound intensity threshold, a wake-up operation is performed.

7. The voice wake-up method according to claim 6, characterized in that: The method further comprises: Obtaining historical sound intensity corresponding to the historical sampling data; Based on the historical sound intensity, the target sound intensity threshold is determined.

8. A voice wake-up device, characterized in that: include: A recognition module, used to recognize historical sampling data collected by a microphone, and obtain recognition result information for indicating whether the historical sampling data includes speech; A determination module, configured to determine a target voice wake-up threshold corresponding to the recognition result information based on a preset voice wake-up threshold; A processing module, configured to perform wake-up processing based on the target voice wake-up threshold; The determination module is specifically used for: In a case where the recognition result information indicates that the historical sampling data includes voice, determining the preset voice wake-up threshold as the target voice wake-up threshold, or determining the preset voice wake-up threshold that has been adjusted higher as the target voice wake-up threshold; In a case where the recognition result information indicates that the historical sampling data does not include voice, the preset voice wake-up threshold that has been adjusted down is determined as the target voice wake-up threshold.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the voice wake-up method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the voice wake-up method according to any one of claims 1 to 7 is implemented.

Images