Wake-up word recognition method, device, electronic device, and storage medium

Abstract

The present disclosure provides a wake-up word recognition method, device, electronic device and storage medium, which belongs to the field of Internet technology. The method includes: recognizing multiple wake-up word voice positive samples by calling a wake-up word recognition model to obtain the near-phonetic word corresponding to each wake-up word; generating multiple wake-up word voice negative samples based on the near-phonetic word corresponding to each wake-up word; training the wake-up word recognition model based on the multiple wake-up word voice positive samples and the multiple wake-up word voice negative samples to obtain the trained wake-up word recognition model. The present disclosure recognizes the wake-up word voice positive samples based on the wake-up word recognition model, and obtains the wake-up word voice negative samples including the near-phonetic word corresponding to the wake-up word. Since the wake-up word voice negative samples contain more near-phonetic words, the trained wake-up word recognition model can learn the voice features of more near-phonetic words, so that it can accurately recognize the wake-up word and its near-phonetic words, achieving the effect of better suppressing false wake-up caused by near-phonetic words.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a wake-up word recognition method, device, electronic device, and storage medium. Background Art

[0002] With the development of artificial intelligence (AI) technology and the widespread adoption of smart voice devices, more and more users are interacting with them through voice messages. During this interaction, the smart device recognizes the user's voice message and, if it detects a wake-up word, controls the device to perform actions such as playing music, adjusting the air conditioning temperature, or setting an alarm.

[0003] At present, the relevant technology mainly adopts the following method to recognize wake-up words: obtain multiple positive wake-up word speech samples containing wake-up words and multiple negative wake-up word speech samples containing non-wake-up words; based on the multiple positive wake-up word speech samples and the multiple negative wake-up word speech samples, train the wake-up word recognition model to obtain the trained wake-up word recognition model; call the trained wake-up word recognition model to recognize the speech information and obtain the recognition result.

[0004] Since the non-wake-up words included in the above-mentioned wake-up word speech negative samples are randomly sampled and do not include near-phonetic words with similar pronunciation to the wake-up words, the trained wake-up word recognition model cannot learn the speech features of near-phonetic words, and the recognition results are less accurate. In actual applications, false wake-up often occurs. Summary of the Invention

[0005] The embodiments of the present disclosure provide a wake-up word recognition method, device, electronic device, and storage medium that can accurately recognize the wake-up word and its near-sounding words, improve the accuracy of the recognition results, and effectively avoid the occurrence of false wake-ups. The technical solution is as follows:

[0006] In a first aspect, a wake-up word recognition method is provided, the method comprising:

[0007] Obtain multiple wake-up word speech positive samples, each of which contains a wake-up word;

[0008] Recognize the multiple wake-up word voice positive samples by calling the wake-up word recognition model to obtain a near-phonetic word corresponding to each wake-up word, where the near-phonetic word is a word with the same or similar pronunciation as the wake-up word;

[0009] Generate multiple wake-up word speech negative samples based on the similar-sounding words corresponding to each wake-up word;

[0010] Based on the multiple positive wake-up word voice samples and the multiple negative wake-up word voice samples, the wake-up word recognition model is trained to obtain a trained wake-up word recognition model, and the trained wake-up word recognition model is used to perform wake-up word recognition on any voice information.

[0011] In another embodiment of the present disclosure, the identifying the plurality of wake-up word voice positive samples by calling the wake-up word recognition model to obtain a near-sounding word corresponding to each wake-up word includes:

[0012] Based on the speech decoding network corresponding to the wake-up word recognition model, speech decoding is performed on the multiple wake-up word speech positive samples to obtain multiple decoding paths corresponding to each wake-up word;

[0013] According to the multiple decoding paths corresponding to each wake-up word and the decoding path corresponding to each non-wake-up word, the near-sounding word corresponding to each wake-up word is obtained.

[0014] In another embodiment of the present disclosure, before performing speech decoding on the multiple wake-up word speech positive samples based on the speech decoding network corresponding to the wake-up word recognition model to obtain multiple decoding paths corresponding to each wake-up word, the method further includes:

[0015] Increasing a parameter value of a path selection parameter of the wake-up word recognition model, wherein the path selection parameter is used to determine multiple decoding paths for each wake-up word in the speech decoding network;

[0016] The speech decoding network corresponding to the wake-up word recognition model performs speech decoding on the multiple wake-up word speech positive samples to obtain multiple decoding paths corresponding to each wake-up word, including:

[0017] Based on the speech decoding network, speech decoding is performed on a positive wake-up word speech sample containing any wake-up word to obtain multiple candidate paths corresponding to the wake-up word and a path score for each candidate path;

[0018] From the multiple candidate paths, obtain the candidate path with the highest path score as the optimal decoding path for the wake-up word;

[0019] For any candidate path other than the optimal decoding path, if the difference between the path score of the optimal decoding path and the path score of the candidate path is less than the parameter value of the adjusted path selection parameter, the candidate path is used as the decoding path of the wake-up word.

[0020] In another embodiment of the present disclosure, obtaining a near-sounding word corresponding to each wake-up word according to a decoding path corresponding to each wake-up word and a decoding path corresponding to each non-wake-up word includes:

[0021] When the decoding path corresponding to any non-wake-up word is the same as any decoding path corresponding to any wake-up word, the non-wake-up word is determined to be a near-phonetic word of the wake-up word.

[0022] In another embodiment of the present disclosure, obtaining a near-sounding word corresponding to each wake-up word according to a decoding path corresponding to each wake-up word and a decoding path corresponding to each non-wake-up word includes:

[0023] When the difference between the path score of the decoding path corresponding to any non-wake-up word and the path score of any decoding path corresponding to any wake-up word is less than a preset threshold, the non-wake-up word is determined to be a near-phonetic word of the wake-up word.

[0024] In another embodiment of the present disclosure, generating multiple wake-up word voice negative samples based on the near-sounding words corresponding to each wake-up word includes:

[0025] Based on the similar-sounding words corresponding to each wake-up word, obtain the negative speech samples of the similar-sounding words corresponding to each wake-up word;

[0026] The near-pronunciation word speech negative sample corresponding to each wake-up and multiple random speech negative samples are combined to form the multiple wake-up word speech negative samples.

[0027] In another embodiment of the present disclosure, obtaining a negative speech sample of the near-pronunciation word corresponding to each wake-up word based on the near-pronunciation word corresponding to each wake-up word includes:

[0028] According to the number of similar-sounding words corresponding to each wake-up word, determine the acquisition ratio of the similar-sounding word speech negative samples corresponding to each wake-up word;

[0029] According to the acquisition ratio of the near-pronunciation word negative speech samples corresponding to each wake-up word and the total number of wake-up word negative speech samples that need to be obtained, the near-pronunciation word negative speech samples corresponding to each wake-up word are obtained.

[0030] In a second aspect, a wake-up word recognition device is provided, the device comprising:

[0031] An acquisition module is used to obtain multiple positive samples of wake-up word speech, each positive sample of wake-up word speech contains a wake-up word;

[0032] The acquisition module is further configured to recognize the plurality of wake-up word voice positive samples by calling a wake-up word recognition model to obtain a near-phonetic word corresponding to each wake-up word, where the near-phonetic word is a word with the same or similar pronunciation as the wake-up word;

[0033] A generation module is used to generate multiple wake-up word speech negative samples based on the similar-sounding words corresponding to each wake-up word;

[0034] A training module is used to train the wake-up word recognition model based on the multiple wake-up word voice positive samples and the multiple wake-up word voice negative samples to obtain a trained wake-up word recognition model, and the trained wake-up word recognition model is used to perform wake-up word recognition on any voice information.

[0035] In another embodiment of the present disclosure, the acquisition module is used to perform speech decoding on the multiple wake-up word speech positive samples based on the speech decoding network corresponding to the wake-up word recognition model to obtain multiple decoding paths corresponding to each wake-up word; perform speech decoding on multiple non-wake-up word speech based on the speech decoding network to obtain the decoding path corresponding to each non-wake-up word; and obtain the near-sounding word corresponding to each wake-up word according to the multiple decoding paths corresponding to each wake-up word and the decoding path corresponding to each non-wake-up word.

[0036] In another embodiment of the present disclosure, the apparatus further comprises:

[0037] an adjustment module, configured to increase a parameter value of a path selection parameter of the wake-up word recognition model, wherein the path selection parameter is used to determine a decoding path for each wake-up word in the speech decoding network;

[0038] The acquisition module is used to perform speech decoding on a positive wake-up word speech sample containing any wake-up word based on the speech decoding network to obtain multiple candidate paths corresponding to the wake-up word and a path score of each candidate path; from the multiple candidate paths, obtain the candidate path with the highest path score as the optimal decoding path for the wake-up word; for any candidate path other than the optimal decoding path, if the difference between the path score of the optimal decoding path and the path score of the candidate path is less than the parameter value of the adjusted path selection parameter, then the candidate path is used as the decoding path for the wake-up word.

[0039] In another embodiment of the present disclosure, the acquisition module is configured to determine the non-wake-up word as a near-phonetic word of the wake-up word when the decoding path corresponding to any non-wake-up word is the same as any decoding path corresponding to any wake-up word.

[0040] In another embodiment of the present disclosure, the acquisition module is used to determine the non-wake-up word as a near-phonetic word of the wake-up word when the difference between the path score of the decoding path corresponding to any non-wake-up word and the path score of any decoding path corresponding to any wake-up word is less than a preset threshold.

[0041] In another embodiment of the present disclosure, the generation module is used to obtain the near-phonetic word speech negative sample corresponding to each wake-up word based on the near-phonetic word corresponding to each wake-up word; and combine the near-phonetic word speech negative sample corresponding to each wake-up word and multiple random speech negative samples into the multiple wake-up word speech negative samples.

[0042] In another embodiment of the present disclosure, the generation module is used to determine the acquisition ratio of the near-phonetic word negative speech samples corresponding to each wake-up word based on the number of near-phonetic words corresponding to each wake-up word; and obtain the near-phonetic word negative speech samples corresponding to each wake-up word based on the acquisition ratio of the near-phonetic word negative speech samples corresponding to each wake-up word and the total number of wake-up word negative speech samples that need to be obtained.

[0043] In a third aspect, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one program code, and the at least one program code is loaded and executed by the processor to implement the wake-up word recognition method as described in the first aspect.

[0044] In a fourth aspect, a computer-readable storage medium is provided, wherein at least one program code is stored in the storage medium, and the at least one program code is loaded and executed by a processor to implement the wake-up word recognition method as described in the first aspect.

[0045] In a fifth aspect, a computer program product includes a computer program code, the computer program code is stored in a computer-readable storage medium, a processor of an electronic device reads the computer program code from the computer-readable storage medium, and the processor executes the computer program code, so that the electronic device performs the wake-up word recognition method described in the first aspect.

[0046] The technical solutions provided by the embodiments of the present disclosure have the following beneficial effects:

[0047] The wake-up word speech positive samples are identified based on the wake-up word recognition model, and the wake-up word speech negative samples including the near-phonetic words corresponding to the wake-up word are obtained. Since the wake-up word speech negative samples contain more near-phonetic words, the wake-up word recognition model trained based on the wake-up word speech negative samples can learn more speech features of near-phonetic words. When the speech information is recognized based on the trained wake-up word recognition model, the wake-up word and the near-phonetic words of the wake-up word can be accurately identified, thereby avoiding false awakenings due to near-phonetic words, and achieving the effect of better suppressing false awakenings due to near-phonetic words. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0049] Figure 1This is a flowchart of a wake-up word recognition method provided by an embodiment of the present disclosure;

[0050] Figure 2 is a flowchart of another wake-up word recognition method provided by an embodiment of the present disclosure;

[0051] Figure 3 is a structural diagram of a wake-up word recognition device provided by an embodiment of the present disclosure;

[0052] Figure 4 A structural block diagram of an electronic device provided by an exemplary embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0053] In order to make the objectives, technical solutions and advantages of the present disclosure more clear, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.

[0054] It should be understood that the terms "each," "plurality," and "any" used in the embodiments of the present disclosure include two or more, each refers to each of the corresponding plurality, and any refers to any one of the corresponding plurality. For example, if a plurality of words includes 10 words, each refers to each of the 10 words, and any refers to any one of the 10 words.

[0055] Before implementing the embodiments of the present disclosure, the terms involved in and the technologies adopted in the embodiments of the present disclosure are first introduced.

[0056] Voice wake-up: When a smart voice device is in standby mode, power consumption is low. By setting a wake-up word in the smart voice device or control application, the smart voice device will wake up when the user's voice message contains the wake-up word, allowing it to enter normal operation and perform the corresponding operation. Many smart voice devices currently have a voice wake-up function, which can be used to unlock the screen or as an auxiliary means of launching applications.

[0057] Speech Technology: includes automatic speech recognition technology, speech synthesis technology, and voiceprint recognition technology. Automatic speech recognition technology is a technology that converts speech into text, and speech synthesis technology is a technology that converts text into speech.

[0058] The main steps of speech recognition include:

[0059] (1) Preprocessing: Process the input original speech information, filter out unimportant information and background noise, and perform relevant transformation processing.

[0060] (2) Feature extraction process: Extract key feature parameters reflecting speech information to form a feature vector sequence, the most commonly used one is the Mel-frequency cepstral coefficient.

[0061] (3) Acoustic model training process: The acoustic model parameters are trained based on the feature parameters in the training speech library. During recognition, the feature parameters of the speech information to be recognized are matched with the acoustic model to obtain the recognition result.

[0062] (4) Language model training process: The training text database is subjected to grammatical and semantic analysis, and a language model is obtained through training based on a statistical model. The language model can effectively combine the knowledge of Chinese grammar and semantics to describe the internal relationship between words, thereby improving the recognition rate and reducing the search scope.

[0063] (5) Speech decoding: For the input speech information, a recognition network is established based on the trained acoustic model, language model and dictionary. The optimal decoding path is found in the speech decoding network according to the search algorithm. The optimal decoding path is the word string that can output the speech information with the highest probability.

[0064] The present disclosure provides a method for identifying a wake-up word. Figure 1 , the method process provided by the embodiment of the present disclosure includes:

[0065] 101. Obtain multiple positive wake-up word speech samples.

[0066] Among them, each wake-up word voice positive sample contains a wake-up word.

[0067] 102. By calling the wake-up word recognition model to recognize multiple wake-up word voice positive samples, obtain the similar-sounding words corresponding to each wake-up word.

[0068] Among them, homophones are words that have the same or similar pronunciation as the wake-up word.

[0069] 103. Based on the similar-sounding words corresponding to each wake-up word, generate multiple wake-up word speech negative samples.

[0070] 104. Based on multiple positive wake-up word speech samples and multiple negative wake-up word speech samples, a wake-up word recognition model is trained to obtain a trained wake-up word recognition model.

[0071] Among them, the trained wake-up word recognition model is used to recognize the wake-up word of any voice information.

[0072] The method provided by the embodiments of the present disclosure recognizes the positive sample of the wake-up word speech based on the wake-up word recognition model, and obtains the negative sample of the wake-up word speech including the near-phonetic words corresponding to the wake-up word. Since the negative sample of the wake-up word speech contains more near-phonetic words, the wake-up word recognition model trained based on the negative sample of the wake-up word speech can learn more speech features of near-phonetic words. When the speech information is recognized based on the trained wake-up word recognition model, the wake-up word and the near-phonetic words of the wake-up word can be accurately identified, thereby avoiding the false awakening of the near-phonetic words, and achieving the effect of better suppressing the false awakening of the near-phonetic words.

[0073] In another embodiment of the present disclosure, a wake-up word recognition model is called to recognize multiple wake-up word speech positive samples to obtain a near-sounding word corresponding to each wake-up word, including:

[0074] Based on the speech decoding network corresponding to the wake-up word recognition model, multiple wake-up word speech positive samples are decoded to obtain multiple decoding paths corresponding to each wake-up word;

[0075] Based on the speech decoding network, multiple non-wake-up word voices are decoded to obtain the decoding path corresponding to each non-wake-up word;

[0076] According to the multiple decoding paths corresponding to each wake-up word and the decoding path corresponding to each non-wake-up word, the near-sounding word corresponding to each wake-up word is obtained.

[0077] In another embodiment of the present disclosure, based on the speech decoding network corresponding to the wake-up word recognition model, before performing speech decoding on multiple wake-up word speech positive samples to obtain multiple decoding paths corresponding to each wake-up word, the method further includes:

[0078] Increase the value of the path selection parameter of the wake-up word recognition model. This path selection parameter is used to determine the multiple decoding paths for each wake-up word in the speech decoding network.

[0079] Based on the speech decoding network corresponding to the wake-up word recognition model, multiple wake-up word speech positive samples are decoded to obtain multiple decoding paths corresponding to each wake-up word, including:

[0080] Based on the speech decoding network, during the process of speech decoding of the wake-up word speech positive sample containing any wake-up word, multiple candidate paths corresponding to the wake-up word and the path score of each candidate path are obtained;

[0081] From multiple candidate paths, the candidate path with the highest path score is obtained as the optimal decoding path for the wake-up word;

[0082] For any candidate path other than the best decoding path, if the difference between the path score of the best decoding path and the path score of the candidate path is less than the parameter value of the adjusted path selection parameter, the candidate path is used as the decoding path of the wake-up word.

[0083] In another embodiment of the present disclosure, obtaining a near-sounding word corresponding to each wake-up word according to a decoding path corresponding to each wake-up word and a decoding path corresponding to each non-wake-up word includes:

[0084] When the decoding path corresponding to any non-wake-up word is the same as the decoding path corresponding to any wake-up word, the non-wake-up word is determined to be a near-phonetic word of the wake-up word.

[0085] In another embodiment of the present disclosure, obtaining a similar word corresponding to each wake-up word according to multiple decoding paths corresponding to each wake-up word and multiple decoding paths corresponding to each non-wake-up word includes:

[0086] When the difference between the path score of the decoding path corresponding to any non-wake-up word and the path score of any decoding path corresponding to any wake-up word is less than a preset threshold, the non-wake-up word is determined to be a near-phonetic word of the wake-up word.

[0087] In another embodiment of the present disclosure, multiple wake-up word voice negative samples are generated based on the similar-sounding words corresponding to each wake-up word, including:

[0088] Based on the similar-sounding words corresponding to each wake-up word, obtain the negative speech samples of the similar-sounding words corresponding to each wake-up word;

[0089] The near-sounding word speech negative samples corresponding to each wake-up and multiple random speech negative samples are combined into multiple wake-up word speech negative samples.

[0090] In another embodiment of the present disclosure, based on the near-pronunciation words corresponding to each wake-up word, obtaining a negative speech sample of the near-pronunciation words corresponding to each wake-up word includes:

[0091] According to the number of similar-sounding words corresponding to each wake-up word, determine the acquisition ratio of the similar-sounding word speech negative samples corresponding to each wake-up word;

[0092] According to the acquisition ratio of the near-pronunciation word negative speech samples corresponding to each wake-up word and the total number of wake-up word negative speech samples that need to be obtained, the near-pronunciation word negative speech samples corresponding to each wake-up word are obtained.

[0093] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present disclosure, and will not be described in detail here.

[0094] The present disclosure provides a wake-up word recognition method. Taking an electronic device as an example, the electronic device has a certain computing capability and can be a smart phone, smart speaker, smart elevator, smart air conditioner, laptop computer, desktop computer, etc. Figure 2 , the method process provided by the embodiment of the present disclosure includes:

[0095] 201. The electronic device obtains multiple positive wake-up word voice samples.

[0096] With the development of artificial intelligence (AI) technology, more and more intelligent voice devices are equipped with voice wake-up functions. Because different intelligent voice devices are designed to perform different functions, the corresponding wake-up words for these functions are also different. Therefore, to enable intelligent voice devices to perform their functions, wake-up words can be set according to the functions of the intelligent voice devices. For example, the functions of a smart speaker are mainly to turn on the speaker, turn off the speaker, pause the currently playing audio file, play the previous audio file, play the next audio file, increase the volume, and decrease the volume. Therefore, the wake-up words for a smart speaker may be "Turn on speaker," "Turn off speaker," "Pause speaker," "Previous song," "Next song," "Turn up volume," "Turn down volume," etc. For another example, the functions of a smart air conditioner are mainly to turn on and off the air conditioner, and adjust the temperature. Therefore, the wake-up words for a smart air conditioner may be "Turn on air conditioner," "Turn off air conditioner," "Set to ** degree," etc.

[0097] Based on multiple wake-up words pre-set for the intelligent voice device, the electronic device obtains the audio files recorded by the user for each wake-up word and obtains multiple positive wake-up word speech samples. Each positive wake-up word speech sample contains a wake-up word. Each wake-up word corresponds to multiple positive wake-up word speech samples to improve the accuracy of each wake-up word recognition.

[0098] 202. The electronic device performs speech decoding on multiple positive wake-up word speech samples based on the speech decoding network corresponding to the wake-up word recognition model to obtain multiple decoding paths corresponding to each wake-up word.

[0099] Among them, the wake-up word recognition model is used to identify the wake-up word from the voice information, and the wake-up word recognition model includes but is not limited to at least one acoustic model such as GMM (Gauss of mixture models) -HMM (Hidden Markov Model), DNN (Deep Neural Networks), CNN (Convolutional Neural Network), LSTM (Long Short-Term Memory). When the voice information to be recognized is input into the wake-up word recognition model, the wake-up word recognition model outputs the probability score of the voice information containing a certain wake-up word. When the probability score is greater than the preset score threshold, it is determined that the voice information contains the wake-up word, thereby controlling the intelligent voice device to perform the operation corresponding to the wake-up word. Among them, the preset score threshold can be 2 points, 2.5 points, etc., and the embodiment of the present disclosure does not make a specific limitation on the preset score threshold.

[0100] In the field of speech technology, the wake-up word recognition model corresponds to a speech decoding network, which includes multiple state nodes, including a state node for the wake-up word. A weight value is set on the edge between the two state nodes, and the weight value represents the posterior probability of jumping from the previous state node to the next state node under the premise of successfully decoding the wake-up word. The speech decoding network can be WFST (Weighted Finite-State Transducers, weighted finite state machine converter) or the like. The speech decoding process is actually a process of searching for state nodes in the speech decoding network where the wake-up word may be, and then connecting the searched state nodes in the search order to form a decoding path for the wake-up word. In order to improve the accuracy of speech decoding of the wake-up word, when searching for the decoding path of the wake-up word, the search can be based on the path selection parameter. Among them, the path selection parameter is used to determine the decoding path of the wake-up word in the speech decoding network. The path selection parameter can be represented by beam, and the decoding path can be represented by N-best. Generally, the larger the parameter value of the path selection parameter, the more decoding paths are searched for the wake-up word in the speech decoding network, and the lower the speech decoding accuracy of the wake-up word is accordingly; the smaller the parameter value of the path selection parameter, the fewer decoding paths are searched for the wake-up word in the speech decoding network, and the higher the speech decoding accuracy of the wake-up word is accordingly. In the method provided by the embodiment of the present disclosure, when adjusting the path selection parameter, the electronic device needs to take into account both accuracy and the number of decoding paths. Under the premise of ensuring accuracy, the parameter value of the path selection parameter is adjusted as large as possible to obtain more decoding paths for the wake-up word, thereby obtaining more near-phonetic words based on the obtained decoding paths, so as to improve the recognition ability of the trained wake-up word recognition model for the wake-up word and its near-phonetic words.

[0101] In an embodiment of the present disclosure, based on the determined parameter value of the path selection parameter and the speech decoding network corresponding to the wake-up word recognition model, the electronic device may use the following method when performing speech decoding on multiple wake-up word speech positive samples:

[0102] 2021. Based on the speech decoding network, the electronic device performs speech decoding on the positive wake-up word speech sample containing any wake-up word to obtain multiple paths corresponding to the wake-up word and the path score of each candidate path.

[0103] For a positive sample of wake-up word speech containing any wake-up word, the electronic device performs speech decoding on the positive sample of wake-up word speech containing the wake-up word based on the speech decoding network corresponding to the wake-up word recognition model. During the speech decoding process, the electronic device uses a token (i.e., a token) to record the weight value and information of a certain state at a certain moment in the wake-up word decoding process. Starting from the initial state of the weighted finite state machine, the token is transferred along the edge with a direction. The change of state during the transfer process can be reflected by the change of the input symbol. In the process of state transfer from the initial state to the terminal state, a series of paths consisting of states and edges are recorded in the token. According to the states recorded in multiple tokens, the electronic device connects each state node in the order of search time to obtain multiple candidate paths for at least the wake-up word, and obtains the weight value of each edge in each candidate path, and then obtains the path score of each candidate path by accumulating or multiplying the weight values of all edges in each candidate path.

[0104] 2022. The electronic device obtains the candidate path with the highest path score from multiple candidate paths as the optimal decoding path for the wake-up word

[0105] 2023. For any candidate path other than the optimal decoding path, if the difference between the path score of the optimal decoding path and the path score of the candidate path is less than the parameter value of the adjusted path selection parameter, the electronic device uses the candidate path as the decoding path of the wake-up word.

[0106] Based on the path score of the determined optimal decoding path, for any candidate path other than the optimal decoding path, the electronic device calculates the difference between the path score of the optimal decoding path and the path score of the candidate path, and then compares the difference with the parameter value of the adjusted path selection parameter. When the difference is less than the parameter value of the adjusted path selection parameter, the electronic device uses the candidate path as the decoding path for the wake-up word.

[0107] It should be noted that the above description uses any candidate path other than the optimal decoding path as an example, and the above method can be used to judge other candidate paths. When all candidate paths other than the optimal decoding path have completed the above judgment process, the electronic device obtains multiple decoding paths corresponding to the wake-up word, including the optimal decoding path.

[0108] 203. The electronic device performs speech decoding on multiple non-wake-up word voices based on a speech decoding network to obtain a decoding path corresponding to each non-wake-up word.

[0109] The electronic device performs speech decoding on multiple non-wake-up word voices based on the speech decoding network corresponding to the wake-up word recognition model. During the speech decoding process of the non-wake-up word voices, the electronic device obtains a decoding path for each non-wake-up word voice. The number of decoding paths can be one or more.

[0110] 204. The electronic device obtains a near-sounding word corresponding to each wake-up word based on the multiple decoding paths corresponding to each wake-up word and the decoding path corresponding to each non-wake-up word.

[0111] Among them, near-phonetic words are words with the same or similar pronunciation as the wake-up word. For example, the wake-up word is "hello", and the near-phonetic words can be "nihao", "nihao", etc. In the field of speech recognition, when the voice information contains near-phonetic words of the wake-up word, false awakening is likely to occur. In order to improve the accuracy of the wake-up word results and reduce the number of false recognitions, the electronic device needs to obtain the near-phonetic words corresponding to each wake-up word. For these near-phonetic words with similar pronunciation to the wake-up word, the electronic device can obtain them according to the multiple decoding paths corresponding to each wake-up word and the decoding path corresponding to each non-wake-up word. The following methods can be used for specific implementation:

[0112] In a possible implementation, when the decoding path corresponding to any non-wake-up word speech is the same as any decoding path of any wake-up word, the electronic device determines the non-wake-up word as a near-sounding word of the wake-up word.

[0113] For any non-wake-up word, the electronic device can match each decoding path corresponding to the non-wake-up word voice with each decoding path of each wake-up word. When any decoding path corresponding to the non-wake-up word voice is the same as any decoding path of any wake-up word, the electronic device determines the non-wake-up word as a near-phonetic word of the wake-up word; the electronic device also obtains the best decoding path with the highest path score from the decoding paths corresponding to the non-wake-up word voice, and then matches the best decoding path of the non-wake-up word with each decoding path of each wake-up word. When the best decoding path corresponding to the non-wake-up word voice is the same as any decoding path of any wake-up word, the electronic device determines the non-wake-up word as a near-phonetic word of the wake-up word.

[0114] In another possible implementation, when the difference between the path score of a decoding path corresponding to any non-wake-up word and the path score of any decoding path corresponding to any wake-up word is less than a preset threshold, the electronic device determines the non-wake-up word as a near-sounding word of the wake-up word. The preset score threshold can be determined based on the electronic device's recognition accuracy requirements for the wake-up word.

[0115] For any non-wake-up word, the electronic device may compare the path score of each decoding path corresponding to the non-wake-up word with the path score of each decoding path corresponding to each wake-up word. When the difference between the path score of any decoding path corresponding to the non-wake-up word and the path score of any decoding path corresponding to any wake-up word is less than a preset threshold, the electronic device determines the non-wake-up word as a near-phonetic word of the wake-up word; the electronic device may also obtain the best decoding path with the highest path score from the decoding path corresponding to the non-wake-up word voice, and then compare the path score of the best decoding path corresponding to the non-wake-up word with the path score of each decoding path corresponding to each wake-up word. When the difference between the path score of the best decoding path corresponding to the non-wake-up word and the path score of any decoding path corresponding to any wake-up word is less than a preset threshold, the electronic device determines the non-wake-up word as a near-phonetic word of the wake-up word.

[0116] 205. Based on the similar-sounding words corresponding to each wake-up word, the electronic device generates multiple wake-up word voice negative samples.

[0117] In an embodiment of the present disclosure, when an electronic device generates multiple wake-up word voice negative samples based on the similar-sounding words corresponding to each wake-up word, the following method can be used:

[0118] 2051. The electronic device obtains a negative speech sample of the similar-pronouncing word corresponding to each wake-up word based on the similar-pronouncing word corresponding to each wake-up word.

[0119] When the electronic device obtains a negative speech sample of a near-pronunciation word corresponding to each wake-up word based on the near-pronunciation word corresponding to each wake-up word, the electronic device may use the following method:

[0120] 20511. The electronic device determines the acquisition ratio of negative speech samples of similar-pronunciation words corresponding to each wake-up word based on the number of similar-pronunciation words corresponding to each wake-up word.

[0121] The electronic device compares the near-phonetic words corresponding to each wake-up word based on the number of near-phonetic words corresponding to each wake-up word, and obtains the acquisition ratio of the near-phonetic word voice negative samples corresponding to each wake-up word. The disclosed embodiment can improve the recognition ability of the near-phonetic words of the wake-up word according to the actual application scenario by calculating the ratio of the number of near-phonetic words corresponding to each wake-up word. For example, if there are a large number of near-phonetic words corresponding to a certain wake-up word, it means that the number of false awakening phenomena based on the wake-up word is large. In this case, more near-phonetic word voice negative samples corresponding to the wake-up word can be selected during model training to improve the recognition ability of the trained wake-up word recognition model for the wake-up word and its near-phonetic words; if there are a small number of near-phonetic words corresponding to a certain wake-up word, it means that the number of false awakening phenomena based on the wake-up word is small. In this case, fewer near-phonetic word voice negative samples corresponding to the wake-up word can be selected during model training to improve the training speed of the model.

[0122] 20512. The electronic device obtains the negative speech samples of the similar-sounding words corresponding to each wake-up word based on the acquisition ratio of the negative speech samples of the similar-sounding words corresponding to each wake-up word and the total number of negative speech samples of the wake-up word that need to be obtained.

[0123] 2052. The electronic device combines the near-pronunciation word voice negative sample corresponding to each wake-up and multiple random voice negative samples into multiple wake-up word voice negative samples.

[0124] 206. The electronic device trains a wake-up word recognition model based on multiple positive wake-up word voice samples and multiple negative wake-up word voice samples to obtain a trained wake-up word recognition model.

[0125] In an embodiment of the present disclosure, each positive wake-up word speech sample corresponds to a wake-up word, and each negative wake-up word speech sample corresponds to a non-wake-up word. The electronic device inputs multiple positive wake-up word speech samples and multiple negative wake-up word speech samples into a wake-up word recognition model, outputs the recognition result of each positive wake-up word speech sample and the recognition result of each negative wake-up word speech sample, inputs the recognition result of each positive wake-up word speech sample and the corresponding wake-up word, and the recognition result of each negative wake-up word speech sample and the corresponding non-wake-up word into a pre-constructed loss function to obtain the function value of the loss function. If the function value is greater than a preset threshold, the parameter value of the wake-up word recognition model is adjusted, and multiple positive wake-up word speech samples and multiple negative wake-up word speech samples are input into the wake-up word recognition model with adjusted parameter values, and then the parameter value of the adjusted wake-up word recognition model is continued to be adjusted based on the output recognition result until the function value of the loss function is less than the preset threshold. The electronic device obtains the parameter value when the function value of the loss function is less than the preset threshold, and uses the wake-up word recognition model corresponding to the parameter value when the function value of the loss function is less than the preset threshold as the trained wake-up word recognition model. The trained wake-up word recognition model is used to perform wake-up word recognition on any voice information.

[0126] Based on the trained wake-up word recognition model, if the electronic device is an intelligent voice device, the electronic device can call the trained wake-up word recognition model to recognize the user's voice information and obtain a recognition result. If it is determined based on the recognition result that the voice information includes the target wake-up word, the electronic device controls the execution of the operation corresponding to the target wake-up word. If the electronic device is not an intelligent voice device, the intelligent voice device can implement the voice wake-up function by loading and running the trained wake-up word recognition model.

[0127] The method provided by the embodiments of the present disclosure recognizes the positive sample of the wake-up word speech based on the wake-up word recognition model, and obtains the negative sample of the wake-up word speech including the near-phonetic words corresponding to the wake-up word. Since the negative sample of the wake-up word speech contains more near-phonetic words, the wake-up word recognition model trained based on the negative sample of the wake-up word speech can learn more speech features of near-phonetic words. When the speech information is recognized based on the trained wake-up word recognition model, the wake-up word and the near-phonetic words of the wake-up word can be accurately identified, thereby avoiding the false awakening of the near-phonetic words, and achieving the effect of better suppressing the false awakening of the near-phonetic words.

[0128] See also Figure 3 , an embodiment of the present disclosure provides a wake-up word recognition device, the device comprising:

[0129] An acquisition module 301 is configured to acquire a plurality of positive wake-up word speech samples, each of which contains a wake-up word.

[0130] The acquisition module 301 is further configured to recognize multiple wake-up word speech positive samples by calling a wake-up word recognition model to obtain a near-phonetic word corresponding to each wake-up word, where the near-phonetic word is a word with the same or similar pronunciation as the wake-up word;

[0131] A generation module 302 is configured to generate a plurality of wake-up word speech negative samples based on the near-sounding words corresponding to each wake-up word;

[0132] The training module 303 is used to train the wake-up word recognition model based on multiple positive wake-up word speech samples and multiple negative wake-up word speech samples to obtain a trained wake-up word recognition model. The trained wake-up word recognition model is used to perform wake-up word recognition on any speech information.

[0133] In another embodiment of the present disclosure, an acquisition module 301 is used to perform speech decoding on multiple wake-up word speech positive samples based on a speech decoding network corresponding to a wake-up word recognition model to obtain multiple decoding paths corresponding to each wake-up word; perform speech decoding on multiple non-wake-up word speech based on the speech decoding network to obtain a decoding path corresponding to each non-wake-up word; and obtain a near-sounding word corresponding to each wake-up word based on the multiple decoding paths corresponding to each wake-up word and the decoding path corresponding to each non-wake-up word.

[0134] In another embodiment of the present disclosure, the apparatus further comprises:

[0135] An adjustment module, configured to increase the value of a path selection parameter of the wake-up word recognition model, where the path selection parameter is used to determine the decoding path of each wake-up word in the speech decoding network;

[0136] The acquisition module 301 is used to perform speech decoding on a positive wake-up word speech sample containing any wake-up word based on a speech decoding network to obtain multiple candidate paths corresponding to the wake-up word and a path score of each candidate path; from the multiple candidate paths, the candidate path with the highest path score is obtained as the optimal decoding path for the wake-up word; for any candidate path other than the optimal decoding path, if the difference between the path score of the optimal decoding path and the path score of the candidate path is less than the parameter value of the adjusted path selection parameter, the candidate path is used as the decoding path for the wake-up word.

[0137] In another embodiment of the present disclosure, the acquisition module 301 is configured to determine the non-wake-up word as a near-phonetic word of the wake-up word when the decoding path corresponding to any non-wake-up word is the same as any decoding path corresponding to any wake-up word.

[0138] In another embodiment of the present disclosure, the acquisition module 301 is used to determine the non-wake-up word as a near-phonetic word of the wake-up word when the difference between the path score of the decoding path corresponding to any non-wake-up word and the path score of any decoding path corresponding to any wake-up word is less than a preset threshold.

[0139] In another embodiment of the present disclosure, a generation module 302 is used to obtain a negative speech sample of the near-phonetic word corresponding to each wake-up word based on the near-phonetic word corresponding to each wake-up word; and to combine the negative speech sample of the near-phonetic word corresponding to each wake-up word and multiple random speech negative samples into multiple wake-up word speech negative samples.

[0140] In another embodiment of the present disclosure, a generation module 302 is used to determine the acquisition ratio of the near-phonetic word negative speech samples corresponding to each wake-up word based on the number of near-phonetic words corresponding to each wake-up word; and to obtain the near-phonetic word negative speech samples corresponding to each wake-up word based on the acquisition ratio of the near-phonetic word negative speech samples corresponding to each wake-up word and the total number of wake-up word negative speech samples that need to be obtained.

[0141] To sum up, the device provided by the embodiments of the present disclosure recognizes the positive sample of wake-up word speech based on the wake-up word recognition model, and obtains the negative sample of wake-up word speech including the near-phonetic words corresponding to the wake-up word. Since the negative sample of wake-up word speech contains more near-phonetic words, the wake-up word recognition model trained based on the negative sample of wake-up word speech can learn more speech features of near-phonetic words. When the speech information is recognized based on the trained wake-up word recognition model, the wake-up word and the near-phonetic words of the wake-up word can be accurately identified, thereby avoiding the false awakening of near-phonetic words, and achieving the effect of better suppressing the false awakening of near-phonetic words.

[0142] Figure 4 FIG. 4 is a block diagram of an electronic device 400 according to an exemplary embodiment of the present disclosure. Generally, the device 400 includes a processor 401 and a memory 402 .

[0143] The processor 401 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 401 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 401 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 401 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 401 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0144] The memory 402 may include one or more computer-readable storage media, which may be non-transitory. The memory 402 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 402 is used to store at least one instruction, which is used to be executed by the processor 401 to implement the method for wake-up word recognition provided in the method embodiment of the present disclosure.

[0145] In some embodiments, electronic device 400 may optionally include a peripheral device interface 403 and at least one peripheral device. Processor 401, memory 402, and peripheral device interface 403 may be connected via a bus or signal lines. Each peripheral device may be connected to peripheral device interface 403 via a bus, signal lines, or circuit boards. Specifically, the peripheral device includes a power supply 404.

[0146] The peripheral device interface 403 can be used to connect at least one I / O (Input / Output)-related peripheral device to the processor 401 and the memory 402. In some embodiments, the processor 401, the memory 402, and the peripheral device interface 403 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 401, the memory 402, and the peripheral device interface 403 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0147] Power supply 404 is used to power the various components of electronic device 400. Power supply 404 can be AC power, DC power, a disposable battery, or a rechargeable battery. When power supply 404 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged via a wired line, while a wireless rechargeable battery is a battery that is charged via a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0148] Those skilled in the art will understand that Figure 4 The structure shown in the figure does not constitute a limitation on the electronic device 400, and the electronic device 400 may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.

[0149] In an exemplary embodiment, a computer-readable storage medium including instructions is further provided, such as a memory including instructions, and the instructions can be executed by a processor of the electronic device 400 to perform the above-mentioned wake-up word recognition method. Optionally, the storage medium can be a non-transitory computer-readable storage medium, 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.

[0150] The electronic device provided by the embodiment of the present disclosure recognizes the positive sample of the wake-up word speech based on the wake-up word recognition model, and obtains the negative sample of the wake-up word speech including the near-phonetic words corresponding to the wake-up word. Since the negative sample of the wake-up word speech contains more near-phonetic words, the wake-up word recognition model trained based on the negative sample of the wake-up word speech can learn more speech features of near-phonetic words. When the speech information is recognized based on the trained wake-up word recognition model, the wake-up word and the near-phonetic words of the wake-up word can be accurately identified, thereby avoiding the false awakening of the near-phonetic words, and achieving the effect of better suppressing the false awakening of the near-phonetic words.

[0151] An embodiment of the present disclosure provides a computer-readable storage medium, wherein at least one program code is stored in the storage medium, and the at least one program code is loaded and executed by a processor to implement a wake-up word recognition method. The computer-readable storage medium may be non-transitory. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

[0152] The computer-readable storage medium provided by the embodiments of the present disclosure recognizes the positive wake-up word speech samples based on the wake-up word recognition model, and obtains the negative wake-up word speech samples including the near-phonetic words corresponding to the wake-up word. Since the negative wake-up word speech samples contain more near-phonetic words, the wake-up word recognition model trained based on the negative wake-up word speech samples can learn more speech features of near-phonetic words. When the speech information is recognized based on the trained wake-up word recognition model, the wake-up word and the near-phonetic words of the wake-up word can be accurately identified, thereby avoiding false awakenings due to near-phonetic words, and achieving the effect of better suppressing false awakenings due to near-phonetic words.

[0153] An embodiment of the present disclosure provides a computer program product, which includes a computer program code, which is stored in a computer-readable storage medium. A processor of an electronic device reads the computer program code from the computer-readable storage medium, and the processor executes the computer program code, so that the electronic device performs a wake-up word recognition method.

[0154] The computer program product provided by the embodiments of the present disclosure recognizes the positive wake-up word speech samples based on the wake-up word recognition model, and obtains the negative wake-up word speech samples including the near-phonetic words corresponding to the wake-up word. Since the negative wake-up word speech samples contain more near-phonetic words, the wake-up word recognition model trained based on the negative wake-up word speech samples can learn more speech features of near-phonetic words. When the speech information is recognized based on the trained wake-up word recognition model, the wake-up word and the near-phonetic words of the wake-up word can be accurately identified, thereby avoiding false awakenings due to near-phonetic words, and achieving the effect of better suppressing false awakenings due to near-phonetic words.

[0155] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0156] The above description is merely an optional embodiment of the present disclosure and is not intended to limit the present disclosure. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present disclosure shall be included in the scope of protection of the present disclosure.

Claims

1. A wake-up word recognition method, characterized in that: The method comprises: Obtain multiple wake-up word speech positive samples, each of which contains a wake-up word; Based on the speech decoding network corresponding to the wake-up word recognition model, speech decoding is performed on the multiple wake-up word speech positive samples to obtain multiple decoding paths corresponding to each wake-up word; Based on the speech decoding network, speech decoding is performed on multiple non-wake-up word speech to obtain a decoding path corresponding to each non-wake-up word; Obtain, based on the multiple decoding paths corresponding to each wake-up word and the decoding path corresponding to each non-wake-up word, a near-phonetic word corresponding to each wake-up word, where the near-phonetic word is a word with the same or similar pronunciation as the wake-up word, and the decoding path of the near-phonetic word corresponding to each wake-up word is the same as any decoding path corresponding to the wake-up word, or the difference between the path score of the decoding path of the near-phonetic word corresponding to each wake-up word and the path score of any decoding path corresponding to the wake-up word is less than a preset threshold; Generate multiple wake-up word speech negative samples based on the similar-sounding words corresponding to each wake-up word; Based on the multiple positive wake-up word voice samples and the multiple negative wake-up word voice samples, the wake-up word recognition model is trained to obtain a trained wake-up word recognition model, and the trained wake-up word recognition model is used to perform wake-up word recognition on any voice information.

2. The method according to claim 1, characterized in that Before performing speech decoding on the plurality of positive wake-up word speech samples based on the speech decoding network corresponding to the wake-up word recognition model to obtain a plurality of decoding paths corresponding to each wake-up word, the method further includes: Increasing a parameter value of a path selection parameter of the wake-up word recognition model, wherein the path selection parameter is used to determine a decoding path for each wake-up word in the speech decoding network; The speech decoding network corresponding to the wake-up word recognition model performs speech decoding on the multiple wake-up word speech positive samples to obtain multiple decoding paths corresponding to each wake-up word, including: Based on the speech decoding network, speech decoding is performed on a positive wake-up word speech sample containing any wake-up word to obtain multiple candidate paths corresponding to the wake-up word and a path score for each candidate path; From the multiple candidate paths, obtain the candidate path with the highest path score as the optimal decoding path for the wake-up word; For any candidate path other than the optimal decoding path, if the difference between the path score of the optimal decoding path and the path score of the candidate path is less than the parameter value of the adjusted path selection parameter, the candidate path is used as the decoding path of the wake-up word.

3. The method according to claim 1, characterized in that The method generates multiple wake-up word voice negative samples based on the near-sounding words corresponding to each wake-up word, including: Based on the similar-sounding words corresponding to each wake-up word, obtain the negative speech samples of the similar-sounding words corresponding to each wake-up word; The near-pronunciation word speech negative sample corresponding to each wake-up and multiple random speech negative samples are combined to form the multiple wake-up word speech negative samples.

4. The method according to claim 3, characterized in that The step of obtaining a negative speech sample of a near-phonetic word corresponding to each wake-up word based on the near-phonetic word corresponding to each wake-up word includes: According to the number of similar-sounding words corresponding to each wake-up word, determine the acquisition ratio of the similar-sounding word speech negative samples corresponding to each wake-up word; According to the acquisition ratio of the near-pronunciation word negative speech samples corresponding to each wake-up word and the total number of wake-up word negative speech samples that need to be obtained, the near-pronunciation word negative speech samples corresponding to each wake-up word are obtained.

5. A wake-up word recognition device, characterized in that: The device comprises: An acquisition module is used to obtain multiple positive samples of wake-up word speech, each positive sample of wake-up word speech contains a wake-up word; The acquisition module is further configured to perform speech decoding on the multiple wake-up word speech positive samples based on the speech decoding network corresponding to the wake-up word recognition model to obtain multiple decoding paths corresponding to each wake-up word; perform speech decoding on multiple non-wake-up word speech based on the speech decoding network to obtain a decoding path corresponding to each non-wake-up word; obtain a near-phonetic word corresponding to each wake-up word according to the multiple decoding paths corresponding to each wake-up word and the decoding path corresponding to each non-wake-up word, wherein the near-phonetic word is a word with the same or similar pronunciation as the wake-up word, and the decoding path of the near-phonetic word corresponding to each wake-up word is the same as any decoding path corresponding to the wake-up word, or the difference between the path score of the decoding path of the near-phonetic word corresponding to each wake-up word and the path score of any decoding path corresponding to the wake-up word is less than a preset threshold; A generation module is used to generate multiple wake-up word speech negative samples based on the similar-sounding words corresponding to each wake-up word; A training module is used to train the wake-up word recognition model based on the multiple wake-up word voice positive samples and the multiple wake-up word voice negative samples to obtain a trained wake-up word recognition model, and the trained wake-up word recognition model is used to perform wake-up word recognition on any voice information.

6. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein at least one program code is stored in the memory, and the at least one program code is loaded and executed by the processor to implement the wake-up word recognition method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that At least one program code is stored in the storage medium, and the at least one program code is loaded and executed by the processor to implement the wake-up word recognition method according to any one of claims 1 to 4.

8. A computer program product, characterized in that The computer program product includes a computer program code, which is stored in a computer-readable storage medium. A processor of an electronic device reads the computer program code from the computer-readable storage medium, and the processor executes the computer program code, so that the electronic device performs the wake-up word recognition method according to any one of claims 1 to 4.

Images