A device wake-up method and apparatus, a computer device, and a storage medium

Abstract

This application provides a device wake-up method, apparatus, computer device, and storage medium, belonging to the field of speech recognition technology. The method includes: classifying multiple speech frames in an acquired speech signal to obtain multiple classification information, the classification information indicating the probability that each syllable, character, or word from a target phrase is included in the speech frame; determining the end point of the target phrase based on the multiple classification information, the end point indicating the time when the playback of the target phrase in the speech signal ends; and waking up the target device based on the end point of the target phrase. This technical solution can determine the time when the playback of the target phrase ends in the speech signal and finally wake up the target device at the time when the playback of the target phrase ends, ensuring that the target device is only woken up when the target phrase is completely detected, avoiding false wake-ups and improving the accuracy of wake-up.

Description

Technical Field

[0001] This application relates to the field of speech recognition technology, and in particular to a device wake-up method, apparatus, computer device, and storage medium. Background Technology

[0002] With the development of speech recognition technology, using devices with voice wake-up functionality has become a popular trend. Users can wake up the device by saying specific words, or wake-up words, thus transitioning it from standby to active mode. Improving the success rate of wake-up is a problem that needs to be solved.

[0003] Currently, the usual method is to directly determine whether the acquired voice signal contains keywords or key syllables from the wake-up word. If keywords or key syllables are detected, the device is woken up to improve the success rate of wake-up.

[0004] The problem with the above technical solution is that the device is woken up when it detects keywords or key syllables, which increases the possibility of the device being woken up falsely and reduces the accuracy of the wake-up. Summary of the Invention

[0005] This application provides a device wake-up method, apparatus, computer device, and storage medium that ensures the target device is only woken up when the target phrase is fully detected, thus avoiding false wake-ups and improving wake-up accuracy. The technical solution is as follows:

[0006] On the one hand, a device wake-up method is provided, the method comprising:

[0007] Multiple speech frames in the acquired speech signal are classified to obtain multiple classification information, which are used to indicate the probability that each syllable, each character, or each word in the target phrase is included in the speech frame;

[0008] Based on the multiple classification information, the tail endpoint of the target phrase is determined, and the tail endpoint is used to indicate the time when the playback of the target phrase in the speech signal ends;

[0009] The target device is woken up based on the tail endpoint of the target phrase.

[0010] In some embodiments, classifying multiple speech frames in the acquired speech signal to obtain multiple classification information includes:

[0011] For any speech frame in the speech signal, feature extraction is performed on the speech frame to obtain the speech features of the speech frame;

[0012] The speech features are classified based on a neural network to obtain the classification information of the speech frame. The neural network is used to classify syllables, characters, or words.

[0013] In some embodiments, the method further includes:

[0014] For a target speech frame, obtain the classification information of the adjacent speech frames that are adjacent to the target speech frame, where the target speech frame is any one of the plurality of speech frames;

[0015] Based on the classification information and smoothing coefficient of the adjacent speech frames, the classification information of the target speech frame is smoothed. The smoothing process is used to make the probability change trend of each syllable, each character or each word of the target phrase in the adjacent speech frames tend to be smooth.

[0016] In some embodiments, determining the tail endpoint of the target phrase based on the plurality of classification information includes:

[0017] Based on the multiple classification information, a first target frame is determined, wherein the first target frame is a speech frame that includes the target character or target word in the target phrase for the first time;

[0018] Based on a rectangular sliding window, multiple first speech frames located after the first target frame are processed to obtain multiple first information. The multiple first information is used to indicate the changing trend of the probability of the target character or target word. The rectangular sliding window has a length of a first number of frames and a sliding step size of one frame.

[0019] Based on the aforementioned first information, the tail endpoint of the target phrase is determined.

[0020] In some embodiments, determining the tail endpoint of the target phrase based on the plurality of first pieces of information includes:

[0021] In response to any adjacent first information, if the probability of the target character or the target word changes from greater than a first threshold to less than a first threshold, a second target frame is determined. The second target frame is the first speech frame that the rectangular sliding window slides through when determining the adjacent first information.

[0022] The start time of the second target frame is determined as the end point of the target phrase.

[0023] In some embodiments, determining the tail endpoint of the target phrase based on the plurality of classification information includes:

[0024] Based on the multiple classification information, a third target frame is determined, which is the first speech frame to include the target syllable in the target phrase;

[0025] Based on a triangular sliding window, multiple second speech frames located after the third target frame are processed to obtain multiple second information. The multiple second information is used to indicate the changing trend of the probability of the target syllable. The length of the triangular sliding window is the second number of frames, and the sliding step is one frame.

[0026] Based on the plurality of second information, the tail endpoint of the target phrase is determined.

[0027] In some embodiments, determining the tail endpoint of the target phrase based on the plurality of second information includes:

[0028] Based on the multiple pieces of second information, determine the change in the probability of the target syllable in two adjacent pieces of second information;

[0029] In response to the change amount being greater than the second threshold for the third consecutive number of times, a fourth target frame is determined. The fourth target frame is the second audio frame that the triangular sliding window slides through when the change amount first exceeds the second threshold.

[0030] The start time of the fourth target frame is determined as the end point of the target phrase.

[0031] In some embodiments, the method further includes:

[0032] Obtain the currently input voice signal;

[0033] The speech signal is processed based on a speech sliding window to obtain the plurality of speech frames. The length of the speech sliding window is a first duration, and the sliding step size is a second duration, wherein the second duration is less than the first duration.

[0034] In some embodiments, the method further includes:

[0035] The target device is woken up when the probability of a wake-up syllable, wake-up word, or wake-up phrase in any of the classification information is greater than the wake-up threshold.

[0036] The step of waking up the target device based on the tail endpoint of the target phrase includes:

[0037] When the target device is awake, a first voice signal is obtained from the voice signal, wherein the first voice signal is the voice signal after the tail end point;

[0038] The first and second speech signals are input into the automatic speech recognition model. The second speech signal is a newly acquired speech signal. The automatic speech recognition model is used to recognize the speech signal as an interactive command.

[0039] On the other hand, a device for waking up a device is provided, the device comprising:

[0040] The classification module is used to classify multiple speech frames in the acquired speech signal to obtain multiple classification information. The classification information is used to indicate the probability that each syllable, each character, or each word in the target phrase is included in the speech frame.

[0041] The determining module is used to determine the tail endpoint of the target phrase based on the multiple classification information, wherein the tail endpoint is used to indicate the time when the playback of the target phrase in the speech signal ends;

[0042] The wake-up module is used to wake up the target device based on the tail endpoint of the target phrase.

[0043] In some embodiments, the classification module is used to extract features from any speech frame in the speech signal to obtain speech features of the speech frame; and to classify the speech features based on a neural network to obtain classification information of the speech frame, wherein the neural network is used to classify syllables, characters or words.

[0044] In some embodiments, the apparatus further includes:

[0045] The first acquisition module is used to acquire, for a target speech frame, classification information of adjacent speech frames adjacent to the target speech frame, wherein the target speech frame is any one of the plurality of speech frames.

[0046] The smoothing module is used to smooth the classification information of the target speech frame based on the classification information and smoothing coefficient of the adjacent speech frames. The smoothing process is used to make the probability change trend of each syllable, each character or each word of the target phrase in the adjacent speech frames tend to be smooth.

[0047] In some embodiments, the determining module includes:

[0048] The first determining unit is configured to determine a first target frame based on the multiple classification information, wherein the first target frame is a speech frame that includes the target character or target word in the target phrase for the first time;

[0049] The first processing unit is used to process multiple first speech frames located after the first target frame based on a rectangular sliding window to obtain multiple first information. The multiple first information is used to indicate the changing trend of the probability of the target character or target word. The rectangular sliding window has a length of a first number of frames and a sliding step size of one frame.

[0050] The second determining unit is used to determine the tail endpoint of the target phrase based on the plurality of first information.

[0051] In some embodiments, the second determining unit is configured to determine a second target frame in response to the probability of the target character or the target word changing from greater than a first threshold to less than a first threshold in any adjacent first information, wherein the second target frame is the first speech frame that the rectangular sliding window slides through when determining the adjacent first information; and to determine the start time of the second target frame as the end point of the target phrase.

[0052] In some embodiments, the determining module includes:

[0053] The third determining unit is used to determine a third target frame based on the multiple classification information, wherein the third target frame is a speech frame that includes the target syllable in the target phrase for the first time;

[0054] The second processing unit is used to process multiple second speech frames located after the third target frame based on a triangular sliding window to obtain multiple second information. The multiple second information is used to indicate the changing trend of the probability of the target syllable. The length of the triangular sliding window is the second number of frames, and the sliding step is one frame.

[0055] The fourth determining unit is used to determine the tail endpoint of the target phrase based on the plurality of second information.

[0056] In some embodiments, the fourth determining unit is configured to determine the change in probability of the target syllable in two adjacent pieces of second information based on the plurality of second information; in response to the change being greater than a second threshold for a third consecutive number of times, determine a fourth target frame, wherein the fourth target frame is the second speech frame that the triangular sliding window slides through when the change first exceeds the second threshold; and determine the start time of the fourth target frame as the end point of the target phrase.

[0057] In some embodiments, the apparatus further includes:

[0058] The second acquisition module is used to acquire the currently input voice signal;

[0059] The signal processing module is used to process the speech signal based on a speech sliding window to obtain the plurality of speech frames. The length of the speech sliding window is a first duration, and the sliding step size is a second duration, wherein the second duration is less than the first duration.

[0060] In some embodiments, the wake-up module is further configured to wake up the target device in response to a probability that the wake-up syllable, wake-up character, or wake-up word in the target phrase of any classification information is greater than a wake-up threshold; if the target device is already awake, to acquire a first speech signal from the speech signal, wherein the first speech signal is the speech signal after the tail endpoint; and to input the first speech signal and a second speech signal into an automatic speech recognition model, wherein the second speech signal is a newly acquired speech signal, and the automatic speech recognition model is configured to recognize the speech signal as an interactive command.

[0061] On the other hand, a computer device is provided, the computer device including a processor and a memory, the memory being used to store at least one computer program, the at least one computer program being loaded by the processor and executed by the above-described device wake-up method.

[0062] On the other hand, a computer-readable storage medium is provided for storing at least one computer program for executing the above-described device wake-up method.

[0063] On the other hand, a computer program product is provided, which, when executed by a processor, implements the above-described device wake-up method.

[0064] This application provides a device wake-up method. By classifying multiple speech frames in the acquired speech signal, the probability that each speech frame includes each syllable, each character, or each word of the target phrase can be determined. Based on the probability that each speech frame includes each syllable, each character, or each word of the target phrase, the time when the target phrase ends playing in the speech signal can be determined. Finally, the target device is woken up at the time when the target phrase ends playing. This method ensures that the target device is only woken up when the target phrase is completely detected, avoiding false wake-ups and improving the accuracy of wake-up. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 This is an implementation environment for a device wake-up method provided in an embodiment of this application;

[0067] Figure 2 This is a flowchart of a device wake-up method provided according to an embodiment of this application;

[0068] Figure 3 This is a flowchart of another device wake-up method provided according to an embodiment of this application;

[0069] Figure 4 This is a schematic diagram of a rectangular sliding window provided according to an embodiment of this application;

[0070] Figure 5 This is a schematic diagram of a triangular sliding window provided according to an embodiment of this application;

[0071] Figure 6 This is a block diagram of a device wake-up device according to an embodiment of this application;

[0072] Figure 7 This is a block diagram of another device wake-up device provided according to an embodiment of this application;

[0073] Figure 8 This is a structural block diagram of a terminal provided according to an embodiment of this application;

[0074] Figure 9 This is a schematic diagram of the structure of a server according to an embodiment of this application. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0076] In this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0077] In this application, the term "at least one" means one or more, and "multiple" means two or more.

[0078] It should be noted that all information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in this application have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the voice signals involved in this application were all obtained with full authorization.

[0079] The following explains the terms used in this application.

[0080] Mel-Frequency Cepstrum (MFCC) is a linear transformation of the logarithmic energy spectrum based on a nonlinear Mel scale of sound frequencies. The Mel-Frequency Cepstrum coefficients are the coefficients that make up the Mel-Frequency Cepstrum. It is derived from the cepstrum of an audio segment. The difference between a regular cepstrum and a Mel-Frequency Cepstrum is that the frequency bands of the Mel-Frequency Cepstrum are divided at equal intervals on the Mel scale, which more closely approximates the human auditory system than the linearly spaced bands used in a normal logarithmic cepstrum. This nonlinear representation allows for better representation of sound signals in multiple fields. MFCC features are widely used in speech recognition.

[0081] Filter banks (FBank) are one of the methods for extracting speech feature parameters. Due to their unique cepstral-based extraction method, they better align with human auditory principles, making them one of the most common and effective speech feature extraction algorithms. Based on filter banks, FBank feature extraction is essentially MFCC without the final discrete cosine transform (a lossy transform). Compared to MFCC features, FBank features retain more original speech data. FBank features are used in speech recognition.

[0082] The device wake-up method provided in this application can be executed by a computer device. In some embodiments, the computer device is a terminal or a server. The following section first uses a computer device as an example to describe the implementation environment of the device wake-up method provided in this application. Figure 1 This is a schematic diagram illustrating the implementation environment of a device wake-up method according to an embodiment of this application. See also... Figure 1 The implementation environment includes terminal 101 and server 102.

[0083] Terminal 101 and server 102 can be connected directly or indirectly via wired or wireless communication, and this application does not impose any restrictions on this.

[0084] In some embodiments, terminal 101 is a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited thereto. Terminal 101 has applications that support voice recognition installed and running.

[0085] In some embodiments, the server 102 is an independent physical server, or can also be a server cluster or a distributed system composed of multiple physical servers, and can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The server 102 is used to provide background services for application programs that support virtual scenarios. In some embodiments, the server 102 undertakes the main computing work, and the terminal 101 undertakes the secondary computing work; or, the server 102 undertakes the secondary computing work, and the terminal 101 undertakes the main computing work; or, the server 102 and the terminal 101 adopt a distributed computing architecture for collaborative computing.

[0086] Figure 2 is a flowchart of a device wake-up method provided according to an embodiment of the present application. As Figure 2 shown, in the embodiment of the present application, taking the execution by the terminal as an example for illustration. The device wake-up method includes the following steps:

[0087] 201. The terminal classifies multiple speech frames in the acquired speech signal to obtain multiple classification information, and the classification information is used to indicate the probability that each syllable, each character, or each word in the target phrase is included in the speech frame.

[0088] In the embodiment of the present application, the speech signal is a speech signal collected by the terminal in real time. The speech signal can be collected by the terminal through an internal speech collection device, or can be obtained by the terminal through an external speech collection device. The embodiment of the present application does not limit this. The speech signal is used to wake up the device. Correspondingly, the speech signal can contain information for waking up the device. The terminal frames the acquired speech signal to obtain multiple speech frames. For any speech frame, the terminal can classify the speech frame according to classification methods such as syllables, characters, or words, and determine the probability that each syllable, each character, or each word in the target phrase is included in the speech frame, so as to obtain the classification information of the speech frame.

[0089] For example, taking the content of the speech signal as "Hello friend" and the classification method as character classification, the classification information can represent the probability of including "你" (nǐ), the probability of including "好" (hǎo), the probability of including "朋" (péng), and the probability of including "友" (yǒu) in the speech frames obtained from the speech signal. Similarly, the classification information can also be used to represent the probability that each syllable or each word in the target phrase is included in the speech frame.

[0090] 202. The terminal determines the end point of the target phrase based on multiple classification information. The end point is used to indicate the time when the playback of the target phrase in the speech signal ends.

[0091] In this embodiment, the terminal can further process the aforementioned multiple classification information to determine the probability of the target syllable, target character, or target word appearing in the target phrase within the speech signal, and then determine the moment when the target phrase ends playback in the speech signal. The target phrase is the wake-up word used to activate the device. Wake-up words are generally designed as four-character phrases; too short a wake-up word can easily lead to false wake-ups, while too long a wake-up word degrades the communication experience. The target syllable can be the last syllable of the wake-up word, the target character can be the last character of the wake-up word, and the target word can be the last word of the wake-up word.

[0092] 203. The terminal wakes up the target device based on the tail endpoint of the target phrase.

[0093] In this embodiment, after determining the end of the target phrase, the terminal can ascertain that the target phrase has finished playing. At this point, a wake-up command is used to wake up the target device. The target device can be the terminal itself or a smart device connected to the terminal. After being woken up, the target device can recognize newly acquired voice signals and execute the recognized interactive commands.

[0094] This application provides a device wake-up method. By classifying multiple speech frames in the acquired speech signal, the syllables, characters, or words included in each speech frame can be determined. Based on the syllables, characters, or words included in each speech frame, the time when the target phrase ends playing in the speech signal can be determined. Finally, the target device is woken up at the time when the target phrase ends playing. This method ensures that the target device is woken up only when the target phrase is completely detected, avoiding false wake-ups and improving the accuracy of wake-up.

[0095] Figure 2 The main flow of the device wake-up method provided in the embodiments of this application is illustrated by way of example. The device wake-up method will be described in detail below based on an application scenario. Figure 3 This is a flowchart of another device wake-up method provided according to an embodiment of this application, such as... Figure 3 As shown, this embodiment of the application will be described using an example executed by a terminal. The device wake-up method includes the following steps:

[0096] 301. The terminal acquires the currently input voice signal.

[0097] In this embodiment, the terminal can acquire the currently input voice signal in real time through an internal voice acquisition device or an external voice acquisition device. Optionally, the voice signal can be real-time voice or historical recording. Optionally, the voice signal may or may not include information for waking up the device; this application does not limit the voice signal. In this embodiment, the terminal can acquire the voice signal in real time and process the acquired voice signal simultaneously.

[0098] 302. The terminal processes the voice signal based on the voice sliding window to obtain multiple voice frames. The length of the voice sliding window is the first duration, and the sliding step size is the second duration. The second duration is less than the first duration.

[0099] In this embodiment, the terminal can segment the aforementioned voice signal into frames based on a voice sliding window. The voice sliding window slides for a second duration each time, and each slide extracts a portion of the voice signal of a first duration as a voice frame, thus obtaining multiple voice frames. For example, the first duration can be 10 milliseconds, 16 milliseconds, or 20 milliseconds, and the second duration can be 5 milliseconds, 10 milliseconds, or 15 milliseconds. This embodiment does not limit the length of the voice sliding window or the step size of its sliding.

[0100] During the processing of speech signals, since the second duration of the speech sliding window is shorter than the first duration, that is, the sliding step size of the speech sliding window is shorter than the length of the speech sliding window, there is an overlap between adjacent speech frames. This overlap ensures that no information is missed when processing speech frames, thus improving the accuracy of speech processing.

[0101] In some embodiments, the terminal may not use a sliding window to process the voice signal, but instead directly divide the voice signal by a first duration, that is, extract a segment of the voice signal as a voice frame for each first duration, resulting in multiple voice frames. This application does not limit the length of each extracted voice frame.

[0102] 303. The terminal classifies multiple speech frames to obtain multiple classification information. The classification information is used to indicate the probability that each syllable, each character, or each word in the target phrase is included in the speech frame.

[0103] In this embodiment, the terminal can classify speech frames using different classification methods, including classification by syllable, classification by character, or classification by word. If the terminal classifies speech frames by syllable, the classification information of the speech frame is used to indicate the probability that the speech frame includes syllables from the target phrase; if the terminal classifies speech frames by character, the classification information of the speech frame is used to indicate the probability that the speech frame includes characters from the target phrase; if the terminal classifies speech frames by word, the classification information of the speech frame is used to indicate the probability that the speech frame includes words from the target phrase.

[0104] In some embodiments, the terminal can classify speech frames using a neural network trained on a target phrase. This neural network is used to classify syllables, characters, or words, determining the probability of each syllable, character, or word in the target phrase included in the speech frame. Correspondingly, for any speech frame in the aforementioned speech signal, the terminal performs feature extraction on the speech frame to obtain its speech features. Then, the terminal classifies the speech features based on the neural network to obtain classification information for the speech frame. This classification information indicates the probability of each syllable, character, or word in the target phrase being included in the speech frame. The speech features can be MFCC features or FBank features. The neural network can be an ANN (Artificial Neural Network), a CNN (Convolutional Neural Network), or other neural networks that can be used to classify syllables, characters, or words. This application does not limit the type of neural network or its operational form. By classifying the aforementioned speech frames using a neural network, further processing of the speech signal is achieved. This enables the determination of the information contained in each speech frame, such as the probability of each syllable, each character, or each word in the target phrase, thereby determining the start and end times of the target phrase.

[0105] In some embodiments, the terminal may perform step 304 to smooth the above multiple classification information, or it may skip step 304 and perform step 305 and subsequent steps, that is, step 304 is an optional step.

[0106] 304. The terminal performs smoothing processing on multiple categories of information.

[0107] In this embodiment, the terminal smooths the aforementioned multiple classification information using a smoothing coefficient. Different smoothing coefficients are set for different classification methods. If the classification method is based on syllables, since each syllable occupies a very short time, a smaller smoothing coefficient can be set to reduce the correction amplitude and avoid wake-up delay issues. If the classification method is based on words, since words have longer pronunciations, a larger smoothing coefficient can be set to eliminate noisy classification by the neural network and improve classification sensitivity.

[0108] In some embodiments, the terminal can smooth the classification information of adjacent speech frames using a first-order smoothing formula. Accordingly, for a target speech frame, the terminal obtains the classification information of adjacent speech frames connected to the target speech frame, and then smooths the classification information of the target speech frame based on the classification information of the adjacent speech frames and the smoothing coefficient. This smoothing process is used to smooth the changing trends of the probabilities of each syllable, each character, or each word of the target phrase in adjacent speech frames. The first-order smoothing formula is shown in formula (1) below.

[0109] X i =a*X i-1 +(1-a)*X i (I)

[0110] Among them, X i X represents the classification information of the i-th speech frame; i-1 represents the classification information of the (i-1)th speech frame, and 'a' represents the smoothing coefficient.

[0111] For example, taking neural network classification by syllables as an example, the neural network classifies the i-th speech frame and obtains the probability of each syllable in the target phrase included in the speech frame. Among them, some syllables have a probability of 0, indicating that the speech frame does not include that syllable. In the classification information of two adjacent speech frames, the probability difference of each syllable is small. The above-mentioned first-order smoothing formula can make the probabilities in the classification information smoother.

[0112] 305. The terminal determines the end point of the target phrase based on multiple classification information. The end point is used to indicate the time when the playback of the target phrase in the speech signal ends.

[0113] In this embodiment, the terminal can determine the end point of the target phrase based on the probability of each syllable, character, or word in the classification information. Optionally, depending on the classification method, the terminal can use two methods to determine the end point of the target phrase.

[0114] Method 1: When the classification method is to classify by character or word. Since the above classification information has been smoothed, there will be a slight lag in the above speech frames. The terminal can process the above multiple classification information through a rectangular sliding window to determine the end point of the target phrase. Correspondingly, the terminal determines the first target frame based on the above multiple classification information. Then, the terminal processes multiple first speech frames located after the first target frame based on the rectangular sliding window to obtain multiple first pieces of information. Finally, the terminal determines the end point of the target phrase based on the multiple first pieces of information. Optionally, in response to the probability of the target character or target word in any adjacent first piece of information changing from being greater than the first threshold to being less than the first threshold, the second target frame is determined, and the start time of the second target frame is determined as the end point of the target phrase. Among them, the first target frame is the speech frame that first includes the target character or target word in the target phrase. The rectangular sliding window has a length of the first number of frames and a sliding step of one frame. The multiple first pieces of information are used to indicate the change trend of the probability of the target character or target word, and this change trend is a trend of rising first and then falling, that is, during the process from the start of playing the target character or target word to the end of playing the target character or target word in the speech signal, the change trend of the probability of the target character or target word in the multiple classification information. The target character is the last character in the target phrase, and the target word is the last word in the target phrase. The first threshold is used to determine whether the target character or target word is the end point of the target phrase, and the embodiment of the present application does not limit the value of the first threshold. The rectangular sliding window can also be called a rectangular filter. By the change trend of the probability of the target character or target word in the target phrase, the end point of the target phrase can be determined more accurately, that is, the terminal determines that the complete target phrase has been played, which can effectively avoid the situation of false wake-up and improve the device wake-up accuracy.

[0115] For example, if the target phrase is "Student A", the target character is "学" (Chinese character for "student"). When the terminal first determines the first target frame where the character "学" first appears based on the above multiple classification information, and then obtains multiple first speech frames after the first target frame. Among the multiple first speech frames after the first target frame and adjacent to the first target frame, the character "学" is included. The terminal determines the change trend of the probability of the character "学" in these multiple first speech frames through the rectangular window. For example, the length of the rectangular sliding window is 10 frames, and it slides 1 frame each time. The rectangular sliding window sums up the probabilities of the character "学" within 10 frames as the window score of the rectangular sliding window, that is, the first piece of information, and uses multiple first pieces of information to represent the change trend of the probability of the character "学". If after a certain slide of the rectangular sliding window, in the previous first piece of information and the next first piece of information, the probability of the character "学" changes from being greater than the first threshold to being less than the first threshold, it means that the frame that has been slid through may include the character "学". At this time, this frame is used as the second target frame, and the start time of the second target frame is determined as the end time of the target phrase playback. Figure 4This is a schematic diagram of a rectangular sliding window provided according to an embodiment of this application. Figure 4 As shown, the rectangle represents a rectangular sliding window with a length of 10 frames. Each time, it slides one frame. The multiple vertical rectangles represent the probabilities of target characters or target words in multiple classification information of multiple first speech frames. The higher the height of the rectangle, the greater the probability of the target character or target word.

[0116] Method 2: When the classification method is based on syllables. Since the classification information is smoothed, there are no sudden jumps, but rather a gradual change. The terminal can process the speech frames using a triangular sliding window to determine the tail endpoint of the target phrase. This triangular sliding window is a right-angled triangle split in the middle, similar to a Mel filter bank, using the left half. Accordingly, the terminal determines the third target frame based on the multiple classification information. Then, based on the triangular sliding window, the terminal processes multiple second speech frames following the third target frame to obtain multiple second information. The terminal determines the tail endpoint of the target phrase based on the multiple second information. Optionally, the terminal determines the change in probability of the target syllable in two adjacent pieces of second information based on the multiple second information. In response to the change being greater than a second threshold for a third consecutive number of times, a fourth target frame is determined, and the start time of the fourth target frame is defined as the tail endpoint of the target phrase. The third target frame is the speech frame that first includes the target syllable in the target phrase. The multiple second information is used to indicate the trend of probability change of the target syllable; see the explanation of Method 1 for details, which will not be repeated here. The length of the triangular sliding window is the second number of frames, and the sliding step is one frame. This second threshold is used to determine whether the target byte is the end point of the target phrase. This embodiment does not limit the value of the second threshold. The third number can be 2 times, 3 times, etc. By understanding the relationship between the change in probability of the third consecutive number of times and the second threshold, the end point of the target phrase can be determined more accurately, meaning the terminal has confirmed that the complete target phrase has been played. This effectively avoids false wake-ups and improves the device wake-up accuracy.

[0117] In some embodiments, the terminal may determine the second information using the following formula (2).

[0118]

[0119] Among them, Score l The second information is represented by: l, which represents the length of the triangular sliding window; f, which represents the f-th frame within the triangular sliding window; s, which represents the maximum value within the triangular sliding window (usually set to 2); the ratio of f to l, which represents the weight of the f-th frame; and the product of this weight and s, which represents the weight of the f-th frame within the triangular sliding window. f F represents the probability of the target syllable in the classification information corresponding to the f-th frame within the triangular sliding window.f It will change with the change of f. Due to the shape of the triangular sliding window, the weight of the second speech frame on the left will be less than the weight of the second speech frame on the right. Therefore, formula (2) improves the speech processing accuracy by reducing, keeping unchanged or increasing the classification information of the second speech frame.

[0120] For example, Figure 5 This is a schematic diagram of a triangular sliding window provided according to an embodiment of this application. Figure 5 As shown, the right-angled triangle represents a triangular sliding window with a length of the second number of frames, sliding one frame at a time. The multiple vertical rectangles represent the probabilities of target syllables in multiple classification information of multiple second speech frames; the higher the height of the rectangle, the greater the probability of the target syllable.

[0121] 306. The terminal wakes up the target device based on the tail endpoint of the target phrase.

[0122] In this embodiment, when the terminal determines the end of the target phrase, it can send a wake-up command to the target device. This wake-up command instructs the target device to transition from standby mode to working mode. The target device may be equipped with an automatic speech recognition model. When the target device is in working mode, this model can recognize the input speech signal and interpret it as an interactive command, which the target device can then execute. By determining the complete target phrase before waking the device, false wake-ups are avoided, improving the accuracy of device wake-up.

[0123] In some embodiments, the terminal may first wake up the target device, but instead of sending a voice signal to the automatic speech recognition model in the target device, it processes the received voice signal based on the target phrase before sending it to the automatic speech recognition model. This is to prevent wake-up words in the voice signal from entering the automatic speech recognition model, which could lead to failure or error in the recognition of interactive commands. Accordingly, in response to the probability that the wake-up syllable, wake-up character, or wake-up word in the target phrase in any classification information is greater than the wake-up threshold, the terminal wakes up the target device. Then, the terminal determines the end point of the target phrase based on the above classification information. When the target device is already awake, a first voice signal is obtained from the voice signal, and the first voice signal and a second voice signal are input into the automatic speech recognition model. The first voice signal is the voice signal after the end point, and the second voice signal is the newly acquired voice signal. The wake-up threshold is used to determine whether to wake up the target device. If the probability of the wake-up syllable, wake-up character, or wake-up word in the above classification information is greater than the wake-up threshold, the terminal determines that the voice signal is playing the target phrase, and wakes up the target device. This application does not limit the wake-up threshold. By first determining whether to wake up the target device and then determining whether the end of the target phrase has been reached, the waiting time for waking up can be reduced and the efficiency of device wake-up can be improved. By inputting the first speech signal after the end of the target phrase and the newly acquired second speech signal into the automatic speech recognition model for speech recognition, the automatic speech recognition model avoids recognizing some syllables, some characters or words in the target phrase as interactive commands, thereby improving the accuracy of device wake-up.

[0124] Figure 6 This is a block diagram of a device wake-up apparatus according to an embodiment of this application. The apparatus is used to perform the steps in the above method, see [link to relevant documentation]. Figure 6 The device includes: a classification module 601, a determination module 602, and a wake-up module 603.

[0125] The classification module 601 is used to classify multiple speech frames in the acquired speech signal to obtain multiple classification information. The classification information is used to indicate the probability that each syllable, each character, or each word in the target phrase is included in the speech frame.

[0126] The determination module 602 is used to determine the tail endpoint of the target phrase based on multiple classification information. The tail endpoint is used to indicate the time when the playback of the target phrase in the speech signal ends.

[0127] Wake-up module 603 is used to wake up the target device based on the tail endpoint of the target phrase.

[0128] In some embodiments, the classification module 601 is used to extract features from any speech frame in the speech signal to obtain the speech features of the speech frame; and to classify the speech features based on a neural network to obtain the classification information of the speech frame, wherein the neural network is used to classify syllables, characters or words.

[0129] In some embodiments, Figure 7 This is a block diagram of another device wake-up device according to an embodiment of this application. See also Figure 7 The device also includes:

[0130] The first acquisition module 604 is used to acquire, for the target speech frame, the classification information of the adjacent speech frames adjacent to the target speech frame;

[0131] The smoothing module 605 is used to smooth the classification information of the target speech frame based on the classification information and smoothing coefficient of adjacent speech frames.

[0132] In some embodiments, see Figure 7 The determination module 602 includes:

[0133] The first determining unit 6021 is used to determine a first target frame based on multiple classification information. The first target frame is a speech frame that includes the target character or target word in the target phrase for the first time.

[0134] The first processing unit 6022 is used to process multiple first speech frames located after the first target frame based on a rectangular sliding window to obtain multiple first information. The multiple first information is used to indicate the changing trend of the probability of the target character or target word. The rectangular sliding window has a length of a first number of frames and a sliding step size of one frame.

[0135] The second determining unit 6023 is used to determine the tail end of the target phrase based on multiple pieces of first information.

[0136] In some embodiments, the second determining unit 6023 is configured to determine a second target frame in response to the probability of a target character or target word in any adjacent first information changing from greater than a first threshold to less than a first threshold. The second target frame is the first speech frame that the rectangular sliding window slides through when determining the adjacent first information. The starting time of the second target frame is determined as the end point of the target phrase.

[0137] In some embodiments, see Figure 7 As shown, module 602 includes:

[0138] The third determining unit 6024 is used to determine the third target frame based on multiple classification information. The third target frame is a speech frame that includes the target syllable in the target phrase for the first time.

[0139] The second processing unit 6025 is used to process multiple second speech frames located after the third target frame based on a triangular sliding window to obtain multiple second information. The multiple second information is used to indicate the changing trend of the probability of the target syllable. The length of the triangular sliding window is the second number of frames, and the sliding step is one frame.

[0140] The fourth determining unit 6026 is used to determine the tail end of the target phrase based on multiple pieces of second information.

[0141] In some embodiments, the fourth determining unit 6026 is configured to determine the change in the probability of a target syllable in two adjacent pieces of second information based on a plurality of second information; in response to the change being greater than a second threshold for a third consecutive number of times, determine a fourth target frame, wherein the fourth target frame is the second speech frame that the triangular sliding window slides through when the change first exceeds the second threshold; and determine the start time of the fourth target frame as the end point of the target phrase.

[0142] In some embodiments, see Figure 7 As shown, the device also includes:

[0143] The second acquisition module 606 is used to acquire the currently input voice signal;

[0144] The signal processing module 607 is used to process the speech signal based on the speech sliding window to obtain multiple speech frames. The length of the speech sliding window is a first duration, and the sliding step size is a second duration, which is less than the first duration.

[0145] In some embodiments, the wake-up module 603 is further configured to wake up the target device in response to the probability that the wake-up syllable, wake-up character, or wake-up word in any target phrase in the classification information is greater than the wake-up threshold; when the target device is already awake, to acquire a first speech signal from the speech signal, wherein the first speech signal is the speech signal after the tail end; and to input the first speech signal and the second speech signal into an automatic speech recognition model, wherein the second speech signal is the newly acquired speech signal, and the automatic speech recognition model is used to recognize the speech signal as an interactive command.

[0146] This application provides a device wake-up method. By classifying multiple speech frames in the acquired speech signal, the probability that each speech frame includes each syllable, each character, or each word of the target phrase can be determined. Based on the probability that each speech frame includes each syllable, each character, or each word of the target phrase, the time when the target phrase ends playing in the speech signal can be determined. Finally, the target device is woken up at the time when the target phrase ends playing. This method ensures that the target device is only woken up when the target phrase is completely detected, avoiding false wake-ups and improving the accuracy of wake-up.

[0147] It should be noted that the device wake-up device provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the device wake-up device and the device wake-up method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0148] In the embodiments of this application, the computer device can be configured as a terminal or a server. When the computer device is configured as a terminal, the terminal can act as the execution subject to implement the technical solutions provided in the embodiments of this application. When the computer device is configured as a server, the server can act as the execution subject to implement the technical solutions provided in the embodiments of this application. Alternatively, the technical solutions provided in this application can be implemented through the interaction between the terminal and the server. The embodiments of this application do not limit this.

[0149] When a computer device is configured as a terminal, Figure 8 This is a structural block diagram of a terminal 800 according to an embodiment of this application. The terminal 800 can be a portable mobile terminal, such as a smartphone, tablet computer, MP3 player (Moving Picture Experts Group Audio Layer III), MP4 player (Moving Picture Experts Group Audio Layer IV), laptop computer, or desktop computer. The terminal 800 may also be referred to as user equipment, portable terminal, laptop terminal, desktop terminal, or other names.

[0150] Typically, terminal 800 includes a processor 801 and a memory 802.

[0151] Processor 801 may include one or more processing cores, such as a quad-core processor or an octa-core processor. Processor 801 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 801 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 801 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 801 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0152] The memory 802 may include one or more computer-readable storage media, which may be non-transitory. The memory 802 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 802 are used to store at least one computer program, which is executed by the processor 801 to implement the device wake-up method provided in the method embodiments of this application.

[0153] In some embodiments, the terminal 800 may also optionally include a peripheral device interface 803 and at least one peripheral device. The processor 801, memory 802, and peripheral device interface 803 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 803 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of the following: a radio frequency circuit 804, a display screen 805, a camera assembly 806, an audio circuit 807, and a power supply 808.

[0154] Peripheral device interface 803 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 801 and memory 802. In some embodiments, processor 801, memory 802 and peripheral device interface 803 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 801, memory 802 and peripheral device interface 803 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0155] The radio frequency (RF) circuit 804 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 804 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 804 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. In some embodiments, the RF circuit 804 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The RF circuit 804 can communicate with other terminals through at least one wireless communication protocol. This wireless communication protocol includes, but is not limited to: the World Wide Web, metropolitan area networks, intranets, various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks, and / or WiFi (Wireless Fidelity) networks. In some embodiments, the RF circuit 804 may also include circuitry related to NFC (Near Field Communication), which is not limited in this application.

[0156] Display screen 805 is used to display a UI (User Interface). This UI may include graphics, text, icons, videos, and any combination thereof. When display screen 805 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 801 for processing. In this case, display screen 805 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, there may be one display screen 805, disposed on the front panel of terminal 800; in other embodiments, there may be at least two display screens, disposed on different surfaces of terminal 800 or in a folded design; in other embodiments, display screen 805 may be a flexible display screen, disposed on a curved or folded surface of terminal 800. Furthermore, display screen 805 may be configured as a non-rectangular irregular shape, i.e., a non-rectangular screen. Display screen 805 may be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0157] The camera assembly 806 is used to acquire images or videos. In some embodiments, the camera assembly 806 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of the terminal, and the rear-facing camera is located on the back of the terminal. In some embodiments, there are at least two rear-facing cameras, which are any one of a main camera, a depth-sensing camera, a wide-angle camera, and a telephoto camera, to achieve background blurring by fusion of the main camera and the depth-sensing camera, panoramic shooting by fusion of the main camera and the wide-angle camera, VR (Virtual Reality) shooting, or other fusion shooting functions. In some embodiments, the camera assembly 806 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash is a combination of a warm-light flash and a cool-light flash, which can be used for light compensation at different color temperatures.

[0158] The audio circuit 807 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, converting the sound waves into electrical signals that are input to the processor 801 for processing, or input to the radio frequency circuit 804 to achieve voice communication. For stereo sound acquisition or noise reduction purposes, multiple microphones may be used, each located at a different part of the terminal 800. The microphone may also be an array microphone or an omnidirectional microphone. The speaker is used to convert the electrical signals from the processor 801 or the radio frequency circuit 804 into sound waves. The speaker may be a conventional diaphragm speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can convert electrical signals not only into audible sound waves but also into inaudible sound waves for purposes such as distance measurement. In some embodiments, the audio circuit 807 may also include a headphone jack.

[0159] Power supply 808 is used to supply power to the various components in terminal 800. Power supply 808 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When power supply 808 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.

[0160] In some embodiments, the terminal 800 further includes one or more sensors 809. The one or more sensors 809 include, but are not limited to, an accelerometer 810, a gyroscope 811, a pressure sensor 812, an optical sensor 813, and a proximity sensor 814.

[0161] Accelerometer 810 can detect the magnitude of acceleration on the three coordinate axes of a coordinate system established by terminal 800. For example, accelerometer 810 can be used to detect the components of gravitational acceleration on the three coordinate axes. Processor 801 can control display screen 805 to display the user interface in either a landscape or portrait view based on the gravitational acceleration signal acquired by accelerometer 810. Accelerometer 810 can also be used for games or for acquiring user motion data.

[0162] The gyroscope sensor 811 can detect the orientation and rotation angle of the terminal 800. The gyroscope sensor 811, in conjunction with the accelerometer sensor 810, can collect 3D motion data from the user on the terminal 800. Based on the data collected by the gyroscope sensor 811, the processor 801 can perform the following functions: motion sensing (e.g., changing the UI based on the user's tilt), image stabilization during shooting, game control, and inertial navigation.

[0163] The pressure sensor 812 can be disposed on the side bezel of the terminal 800 and / or the lower layer of the display screen 805. When the pressure sensor 812 is disposed on the side bezel of the terminal 800, it can detect the user's grip signal on the terminal 800, and the processor 801 can perform left / right hand recognition or quick operation based on the grip signal collected by the pressure sensor 812. When the pressure sensor 812 is disposed on the lower layer of the display screen 805, the processor 801 can control the operable controls on the UI interface based on the user's pressure operation on the display screen 805. The operable controls include at least one of button controls, scroll bar controls, icon controls, and menu controls.

[0164] An optical sensor 813 is used to collect ambient light intensity. In one embodiment, the processor 801 can control the display brightness of the display screen 805 based on the ambient light intensity collected by the optical sensor 813. Specifically, when the ambient light intensity is high, the display brightness of the display screen 805 is increased; when the ambient light intensity is low, the display brightness of the display screen 805 is decreased. In another embodiment, the processor 801 can also dynamically adjust the shooting parameters of the camera assembly 806 based on the ambient light intensity collected by the optical sensor 813.

[0165] The proximity sensor 814, also known as a distance sensor, is typically located on the front panel of the terminal 800. The proximity sensor 814 is used to detect the distance between the user and the front of the terminal 800. In one embodiment, when the proximity sensor 814 detects that the distance between the user and the front of the terminal 800 is gradually decreasing, the processor 801 controls the display screen 805 to switch from a screen-on state to a screen-off state; when the proximity sensor 814 detects that the distance between the user and the front of the terminal 800 is gradually increasing, the processor 801 controls the display screen 805 to switch from a screen-off state to a screen-on state.

[0166] Those skilled in the art will understand that Figure 8 The structure shown does not constitute a limitation on terminal 800 and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0167] When a computer device is configured as a server, Figure 9This is a schematic diagram of a server structure according to an embodiment of this application. The server 900 can vary considerably due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 901 and one or more memories 902. The memory 902 stores at least one computer program, which is loaded and executed by the processor 901 to implement the device wake-up method provided in the above-described method embodiments. Of course, the server may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server may also include other components for implementing device functions, which will not be elaborated here.

[0168] This application also provides a computer-readable storage medium storing at least one computer program. This computer program is loaded and executed by the processor of a computer device to implement the operations performed by the computer device in the device wake-up method of the above embodiments. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0169] In some embodiments, the computer program involved in the present application embodiments may be deployed and executed on a computer device, or executed on multiple computer devices located in one location, or executed on multiple computer devices distributed in multiple locations and interconnected through a communication network. Multiple computer devices distributed in multiple locations and interconnected through a communication network may constitute a blockchain system.

[0170] This application also provides a computer program product including computer program code stored in a computer-readable storage medium. A processor of a computer device reads the computer program code from the computer-readable storage medium and executes the computer program code, causing the computer device to perform the device wake-up methods provided in the various optional implementations described above.

[0171] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0172] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A device wake-up method, characterized in that, The method includes: Acquire the currently input speech signal; process the speech signal based on a speech sliding window to obtain multiple speech frames, wherein the length of the speech sliding window is a first duration and the sliding step size is a second duration, and the second duration is less than the first duration; Multiple speech frames in the acquired speech signal are classified to obtain multiple classification information, which are used to indicate the probability that each syllable, each character, or each word in the target phrase is included in the speech frame; For a target speech frame, based on the classification information of adjacent speech frames, and based on the classification information and smoothing coefficient of the adjacent speech frames, the classification information of the target speech frame is smoothed. The smoothing process is used to smooth the probability variation trend of each syllable, each character, or each word of the target phrase in adjacent speech frames. The target speech frame is any one of the multiple speech frames. Different smoothing coefficients are set for different classification methods. If the classification method is based on syllables, a smaller smoothing coefficient is set; if the classification method is based on words, a larger smoothing coefficient is set. Based on the smoothed classification information, a third target frame is determined, which is the first speech frame to include the target syllable in the target phrase. Based on a triangular sliding window, multiple second speech frames located after the third target frame are processed to obtain multiple second information. The multiple second information is used to indicate the changing trend of the probability of the target syllable. The changing trend is an upward trend followed by a downward trend, which is used to represent the process from the start of the speech signal playing the target syllable to the end of the playing of the target syllable. The length of the triangular sliding window is the second number of frames, and the sliding step is one frame. The triangular sliding window is a right-angled triangle with the left half split in the middle, similar to a Mel filter bank. Based on the multiple second pieces of information, the change in probability of the target syllable in two adjacent pieces of second information is determined; in response to the change being greater than a second threshold for a third consecutive number of times, a fourth target frame is determined, the fourth target frame being the second speech frame that the triangular sliding window slides through when the change first exceeds the second threshold; the start time of the fourth target frame is determined as the end point of the target phrase, the end point being used to indicate the time when the playback of the target phrase in the speech signal ends; The target device is woken up based on the tail endpoint of the target phrase.

2. The method according to claim 1, characterized in that, The process of classifying multiple speech frames in the acquired speech signal yields multiple classification information, including: For any speech frame in the speech signal, feature extraction is performed on the speech frame to obtain the speech features of the speech frame; The speech features are classified based on a neural network to obtain the classification information of the speech frame. The neural network is used to classify syllables, characters, or words.

3. The method according to claim 1, characterized in that, The method further includes: The target device is woken up when the probability of a wake-up syllable, wake-up word, or wake-up phrase in any of the classification information is greater than the wake-up threshold. The step of waking up the target device based on the tail endpoint of the target phrase includes: When the target device is awake, a first voice signal is obtained from the voice signal, wherein the first voice signal is the voice signal after the tail end point; The first and second speech signals are input into the automatic speech recognition model. The second speech signal is a newly acquired speech signal. The automatic speech recognition model is used to recognize the speech signal as an interactive command.

4. A device wake-up device, characterized in that, The device includes: The second acquisition module is used to acquire the currently input voice signal; The signal processing module is used to process the speech signal based on a speech sliding window to obtain multiple speech frames. The length of the speech sliding window is a first duration, and the sliding step size is a second duration, wherein the second duration is less than the first duration. The classification module is used to classify multiple speech frames in the acquired speech signal to obtain multiple classification information. The classification information is used to indicate the probability that each syllable, each character, or each word in the target phrase is included in the speech frame. The device is further configured to, for a target speech frame, perform smoothing processing on the classification information of the target speech frame based on the classification information of adjacent speech frames, and based on the classification information of the adjacent speech frames and a smoothing coefficient. The smoothing processing is used to smooth the probability variation trend of each syllable, each character, or each word of the target phrase in adjacent speech frames. The target speech frame is any one of the plurality of speech frames. Different smoothing coefficients are set for different classification methods; if the classification method is based on syllables, a smaller smoothing coefficient is set; if the classification method is based on words, a larger smoothing coefficient is set. The determination module is used to determine a third target frame based on the smoothed multiple classification information. The third target frame is the speech frame that first includes the target syllable in the target phrase. Based on a triangular sliding window, multiple second speech frames located after the third target frame are processed to obtain multiple second information. The multiple second information is used to indicate the changing trend of the probability of the target syllable. The changing trend is an upward trend followed by a downward trend, which is used to represent the process from the start of the speech signal playing the target syllable to the end of the speech signal playing the target syllable. The length of the triangular sliding window is a second number of frames, and the sliding step is one frame. The triangular sliding window is a right-angled triangle with the left half split in the middle, similar to a Mel filter bank. Based on the multiple second information, the change in the probability of the target syllable in two adjacent second information is determined. In response to the change being greater than a second threshold for a third consecutive number of times, a fourth target frame is determined. The fourth target frame is the second speech frame that the triangular sliding window has slid through when the change is first greater than the second threshold. The start time of the fourth target frame is determined as the end point of the target phrase. The end point is used to indicate the time when the playback of the target phrase in the speech signal ends. The wake-up module is used to wake up the target device based on the tail endpoint of the target phrase.

5. A computer device, characterized in that, The computer device includes a processor and a memory, the memory being used to store at least one computer program, the at least one computer program being loaded by the processor and executed as the device wake-up method according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store at least one computer program for performing the device wake-up method according to any one of claims 1 to 3.

7. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the device wake-up method as described in any one of claims 1 to 3.

Images