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

Abstract

The invention discloses a voice wake-up method, a related device, equipment, a system and a storage medium, and the voice wake-up method comprises the steps: obtaining an audio collection result and a state detection result of each seat; based on the audio acquisition result and the state detection result, determining a target sound area number, and based on the state detection result, extracting a first audio of which the audio channel number is the target sound area number from the audio acquisition result; calling a wake-up engine based on the target sound area number to obtain a first wake-up instance; wherein the actual channel number of the wake-up engine in the first wake-up instance is the target sound area number; and inputting the first audio to the first wake-up instance and running the first audio to obtain a first wake-up result. According to the scheme, on the premise that multi-tone-area awakening is achieved, the awakening accuracy and self-adaptability of multi-tone-area awakening to different driving scenes can be improved as much as possible, and resource consumption of multi-tone-area awakening is reduced as soon as possible.

Description

Technical Field

[0001] The present application relates to the field of audio processing technology, and in particular to a voice wake-up method and related devices, equipment, systems and storage media. Background Art

[0002] With the development of science and technology, voice interaction technology has been gradually applied to many scenarios. Taking the in-vehicle interconnection scenario as an example, users are becoming more and more accustomed to interacting with in-vehicle devices through voice, and thus the requirements and demands for in-vehicle voice interaction systems are increasing. Among them, the voice wake-up function, as the "switch" for interaction between people and car machines, has become a key technology to enhance the interactive experience of drivers and passengers.

[0003] At present, voice wake-up technology is usually based on single-zone wake-up, that is, it mainly supports voice wake-up at the main driver's seat, but it is difficult to cover other positions. In addition, single-zone wake-up is easily disturbed by other passengers and cannot meet the increasingly complex driving scenarios. In view of this, how to improve the wake-up accuracy and adaptability of multi-zone wake-up for different driving scenarios as much as possible under the premise of realizing multi-zone wake-up, and reduce the resource consumption of multi-zone wake-up as soon as possible, has become an urgent problem to be solved. Summary of the invention

[0004] The main technical problem solved by the present application is to provide a voice wake-up method and related devices, equipment, systems and storage media, which can improve the wake-up accuracy and adaptability of multi-zone wake-up to different driving scenarios as much as possible under the premise of realizing multi-zone wake-up, and reduce the resource consumption of multi-zone wake-up as soon as possible.

[0005] In order to solve the above-mentioned technical problems, the first aspect of the present application provides a voice wake-up method, including: obtaining audio collection results and status detection results at each seat; wherein the audio collection results include whether the audio is collected and the original audio when the audio is collected, and the status detection results include whether the seat is in a sitting state; based on the audio collection results and the status detection results, determining the target number of sound zones, and based on the status detection results, extracting from the audio collection results a first audio whose number of audio channels is the target number of sound zones; calling a wake-up engine based on the target number of sound zones to obtain a first wake-up instance; wherein the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones; inputting the first audio into the first wake-up instance and running it to obtain a first wake-up result.

[0006] In order to solve the above technical problems, the second aspect of the present application provides a voice wake-up device, including: a result acquisition module, a determination and extraction module, a first calling module and a first running module, the result acquisition module is used to obtain the audio collection results and state detection results at each seat; wherein the audio collection results include whether the audio is collected and the original audio when the audio is collected, and the state detection results include whether it is in a sitting state; the determination and extraction module is used to determine the target number of sound zones based on the audio collection results and the state detection results, and based on the state detection results, extract the first audio whose audio channel number is the target number of sound zones from the audio collection results; the first calling module is used to call the wake-up engine based on the target number of sound zones to obtain a first wake-up instance; wherein the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones; the first running module is used to input the first audio to the first wake-up instance and run it to obtain a first wake-up result.

[0007] In order to solve the above technical problems, the third aspect of the present application provides an electronic device, which at least includes a memory and a processor coupled to each other, the memory at least stores program instructions, and the processor is used to execute the program instructions to implement the voice wake-up method in the above first aspect.

[0008] In order to solve the above-mentioned technical problems, the fourth aspect of the present application provides a voice wake-up system, which at least includes a vehicle computer, and a microphone and a seat sensor respectively arranged on each seat, the microphone and the seat sensor are respectively communicated with the vehicle computer, and the vehicle computer includes at least the electronic device in the above-mentioned third aspect, the microphone is at least used to obtain the audio collection results at the seat, and the seat sensor is used to obtain the status detection results at the seat.

[0009] In order to solve the above technical problem, the fifth aspect of the present application provides a computer-readable storage medium, which stores program instructions that can be executed by a processor, and the program instructions are used to implement the voice wake-up method of the first aspect above.

[0010] The above scheme obtains the audio collection results and state detection results at each seat, and the audio collection results include whether the audio is collected and the original audio when the audio is collected, and the state detection result package is whether the seat is in the sitting state, and then based on the audio collection results and the state detection results, the target number of sound zones is determined, and based on the state detection result, the first audio with the number of audio channels as the target number of sound zones is extracted from the audio collection results, so as to call the wake-up engine based on the target number of sound zones to obtain the first wake-up instance, and the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones, and then the first audio is input into the first wake-up instance and run to obtain the first wake-up result, so on the one hand, the first wake-up result obtained by calling the wake-up engine is run The wake-up instance can realize multi-zone wake-up. On the other hand, since the number of target zones is adaptively determined according to the audio acquisition results and the status detection results, and the actual number of channels of the wake-up engine in the first wake-up instance is the target zone number, compared with the target zone number set as a fixed zone number, the adaptability of multi-zone wake-up to different driving scenarios can be improved. On the other hand, since the number of audio channels of the first audio is also the target zone number adaptively determined, the interference audio of voice wake-up can be excluded as much as possible, and the wake-up accuracy of multi-zone wake-up for different driving scenarios can be improved as much as possible. Compared with processing the audio signals of all zones during the voice wake-up process, the resource consumption of multi-zone wake-up can be reduced as much as possible. Therefore, under the premise of realizing multi-zone wake-up, the wake-up accuracy and adaptability of multi-zone wake-up for different driving scenarios can be improved as much as possible, and the resource consumption of multi-zone wake-up can be reduced as much as possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a flow chart of an embodiment of the voice wake-up method of the present application; Figure 2 This is a process diagram of an embodiment of the voice wake-up method of the present application; Figure 3 It is a schematic diagram of the framework of an embodiment of the voice wake-up device of the present application; Figure 4 It is a schematic diagram of the framework of an embodiment of the electronic device of the present application; Figure 5 It is a schematic diagram of the framework of an embodiment of the voice wake-up system of the present application; Figure 6 It is a schematic diagram of a framework of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0012] The scheme of the embodiment of the present application is described in detail below in conjunction with the drawings of the specification.

[0013] In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures, interfaces, and technologies are provided to facilitate a thorough understanding of the present application.

[0014] The terms "system" and "network" are often used interchangeably in this article. The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the fragment " / " in this article generally indicates that the associated objects before and after are in an "or" relationship. In addition, "many" in this article means two or more than two.

[0015] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of the voice wake-up method of the present application. Specifically, it may include the following steps: Step S11: Obtain the audio collection results and status detection results at each seat.

[0016] In the embodiment of the present disclosure, the audio collection result includes whether the audio is collected and the original audio when the audio is collected, and the state detection result includes whether the user is in a sitting state.

[0017] In one implementation scenario, as a possible example, a microphone may be provided at each seat, and the microphone is used to obtain the audio collection results at the seat. Taking the in-vehicle scenario as an example, microphones may be provided at each seat such as the main driver's seat, the co-driver's seat, and the back row. In addition, microphones may be provided separately at other locations in the car (e.g., locations where audio signal reflection is easily generated) to obtain echo signals. It should be noted that the microphone can cover all possible sound source locations in the usage scenario (e.g., in the car) and collect clear voice commands. The microphone maliciously adopts advanced acoustic design to optimize its sound pickup performance in a noisy environment to effectively suppress background noise, such as ambient noise inside and outside the car, conversations with other passengers, etc., to ensure accurate recognition of specific user voice commands. The microphone may also have good directivity, and the receiving direction may be electronically adjusted and optimized to enhance voice signals from a specific direction and suppress noise and interference from other directions.

[0018] In one implementation scenario, as a possible example, each seat may be provided with a seat sensor, which is used to obtain the status detection result of the seat. For example, the seat sensor may adopt a thin film contact technology, and its contacts may be evenly distributed on the force-bearing surface of the seat. When external pressure acts on the seat, the seat sensor may generate a trigger signal, thereby obtaining the status detection result.

[0019] Step S12: based on the audio collection result and the state detection result, determine the target number of sound zones, and based on the state detection result, extract the first audio whose number of audio channels is the target number of sound zones from the audio collection result.

[0020] In one implementation scenario, after obtaining the audio collection results and state detection results at each seat, the sound zone detection result at the corresponding seat can be determined based on whether the audio collection results and state detection results at the same seat are consistent in connotation, and the sound zone detection result includes whether the area where the seat is located is used as the sound zone, and then statistics are performed based on the sound zone detection results at the seat to obtain the target number of sound zones. The above method determines whether to use the area where the seat is located as the sound zone by combining whether the audio collection results and state detection results at the same seat are consistent in connotation, and then counts the target number of sound zones, which can improve the accuracy of the number of effective sound zones as much as possible under the premise of adaptively determining the number of effective sound zones.

[0021] In a specific implementation scenario, as a possible scenario, in response to the audio collection result at the same seat including the audio being collected and the state detection result including being in the seated state, it can be determined that the connotations are consistent and the sound zone detection result at the corresponding seat includes taking the area where the corresponding seat is located as the sound zone. Exemplarily, if the audio collection result at the main driver's seat includes the audio being collected and the state detection result includes being in the seated state, it can be determined that the audio collection result at the main driver's seat and the state detection result are consistent in connotation, and it can be determined that the sound zone detection result at the main driver's seat includes taking the area where the main driver is located as the sound zone.

[0022] In a specific implementation scenario, different from the aforementioned situation, in response to the audio collection result at the same seat including that the audio is collected and the state detection result includes that the seat is not in the sitting state, or the audio collection result at the same seat includes that the audio is not collected and the state detection result includes that the seat is in the sitting state, it can be determined that the connotation is inconsistent and the sound zone detection result at the corresponding seat includes not using the area where the corresponding seat is located as the sound zone. Exemplarily, if the audio collection result at the co-pilot seat includes that the audio is collected and the state detection result includes that the seat is not in the sitting state, it can be determined that the audio collection result and the state detection result at the co-pilot seat are inconsistent in connotation, and it can be determined that the sound zone detection result at the co-pilot seat includes not using the area where the co-pilot seat is located as the audio.

[0023] In a specific implementation scenario, after obtaining the sound zone detection results at each seat, the sound zone detection results at each seat can be counted to obtain the target number of sound zones. Specifically, the sound zone detection results can include the total number of seats as sound zones as the target number of sound zones. Exemplarily, if, through statistics, the sound zone detection results at the main driver and the leftmost rear seat are included as sound zones, the target number of sound zones can be determined to be 2. Of course, the above example is only a possible example in the actual application process, and the specific value of the target number of sound zones is not limited here.

[0024] In one implementation scenario, after obtaining the audio collection results and state detection results at each seat, the first audio whose number of audio channels is the target number of sound zones can also be extracted from the audio collection results based on the state detection results. As a possible example, if the state detection result at a certain seat includes being in a sitting state, the original audio in the audio collection result at the seat can be extracted, otherwise the original audio in the audio collection result at the seat can be ignored. Of course, if the state detection result at a certain seat includes being in a sitting state, and the audio collection result at the seat includes not collecting audio, the original audio may not be extracted from the audio collection result at the seat. Based on this, the extracted original audios can be combined to obtain the first audio whose number of audio channels is the target number of sound zones. As another possible example, the original audio in the audio collection result at the seat whose sound zone detection result includes the sound zone can also be extracted to obtain the first audio whose number of audio channels is the target number of sound zones. It should be noted that the above examples are only several possible ways to extract the first audio in the actual application process, and the way to extract the first audio is not limited here.

[0025] In one implementation scenario, in order to assist subsequent voice wake-up, after obtaining the first audio, the first audio may be preprocessed. Exemplarily, the preprocessing may include, but is not limited to: echo cancellation processing, narrow beam algorithm processing, etc., and the specific content of the preprocessing is not limited here. It should be noted that the echo cancellation processing may include, but is not limited to: using an adaptive filter (such as a Wiener filter, etc.) to estimate the echo path and generate an anti-phase signal to offset the echo component in the original signal, thereby achieving the purpose of echo cancellation. In addition, the narrow beam algorithm processing may include, but is not limited to: utilizing the spatial distribution characteristics of multiple microphones to adjust and optimize their receiving directions, thereby enhancing the audio signal from a specific direction and suppressing noise and interference from other directions, so as to effectively improve the strength and clarity of the target signal.

[0026] Step S13: calling the wake-up engine based on the target number of sound zones to obtain a first wake-up instance.

[0027] In the embodiment of the present disclosure, the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones. As a possible example, the wake-up engine can support a maximum number of sound zones, such as the maximum number of sound zones that can be supported can be determined according to the application scenario. For example, in a driving scenario, considering that a general family car has less than 7 seats, it can be determined that the maximum number of sound zones supported is 7. Of course, the maximum number of sound zones in other scenarios can be set by analogy, and examples are not given one by one here. On this basis, the actual number of channels of the wake-up engine can be set to the maximum number of sound zones by default, and when the wake-up engine is called to generate a wake-up instance, the actual number of channels of the wake-up engine can be adjusted to the input parameter "target number of sound zones" according to the input parameter "target number of sound zones" to automatically adjust the number of wake-up calculation paths. On the one hand, it can effectively reduce the overall loss of the voice wake-up system. On the other hand, since the actual number of channels can be adjusted in real time according to the adaptively determined target number of sound zones, it can also effectively reduce problems such as false wake-up and sound source localization errors caused by noise interference.

[0028] In an implementation scenario, as a possible example, a wake-up model may be set in the wake-up engine, and the wake-up engine may be a frame-level wake-up model or an end-to-end wake-up model, etc., and the specific type of the wake-up engine is not limited here. It should be noted that the processing process of the frame-level wake-up model and the end-to-end wake-up model can refer to the following related description, which is not repeated here.

[0029] In another implementation scenario, as another possible example, a plurality of wake-up models may also be set in the wake-up engine, such as but not limited to: a frame-level wake-up model, an end-to-end wake-up model, etc., and the specific types of the plurality of wake-up models are not limited here. It should be noted that, as mentioned above, the processing of the frame-level wake-up model and the end-to-end wake-up model can refer to the following related description, which is not repeated here.

[0030] Step S14: input the first audio to the first wake-up instance and run it to obtain a first wake-up result.

[0031] In the disclosed embodiment, the first audio is used as an input parameter of the first instance. In addition, since running the first instance can uniformly perform synchronous calculations on the first audio whose audio channels are the number of target sound zones, it helps to effectively reduce problems such as asynchrony in sound source localization caused by different threads being deployed on different computing cores, that is, it can ensure the calculation synchronization between various sound zones as much as possible.

[0032] In one implementation scenario, as a possible example, taking the case where the wake-up engine includes multiple wake-up models, in the process of running the first wake-up instance, the input audio of the wake-up engine can be processed separately based on different wake-up models (in this implementation scenario, the input audio of the wake-up engine is the first audio), and the candidate wake-up results output by each wake-up model can be obtained, and then the target wake-up result finally output by the wake-up engine can be obtained by analyzing each candidate wake-up result and the time when the result is obtained (in this implementation scenario, the target wake-up result is the first wake-up result). In the above manner, the wake-up engine sets multiple wake-up models, and analyzes the candidate wake-up results of the input audio and the time when the result is obtained by combining multiple wake-up models, which helps to further improve the accuracy of voice wake-up from the perspective of combining multiple models.

[0033] In a specific implementation scenario, as described above, multiple wake-up models may include a frame-level wake-up model, and the first a posteriori information output by the frame-level wake-up model after processing the input audio may be obtained, and the first a posteriori information may specifically include the first probability values ​​of the audio frames in the input audio belonging to several pronunciation states. It should be noted that the specific number of several pronunciation states may depend on the processing frame rate of the frame-level wake-up model. For example, for a low frame rate model, the specific values ​​of several pronunciation states may be 3003, etc., and the specific values ​​of several pronunciation states are not limited here. Further, taking the input audio containing N audio frames as an example, the first a posteriori information may be expressed as a probability matrix of N*3003, and other cases may be deduced by analogy, and examples are not given one by one here. On this basis, the first a posteriori information may be decoded based on a preset decoding strategy to obtain a candidate wake-up result of the frame-level wake-up model, and the preset decoding strategy may at least include filler-keyword. For ease of understanding, the decoding process of filler-keyword is briefly described below. The main goal of the decoding strategy of filler-keyword is to find the maximum score in the acoustic scores of consecutive frames (such as the aforementioned first probability value). Because each frame is interfered by the score of non-target speech, it is necessary to separate the wake-up score from the mixed score. In this process, the wake-up score of non-target speech is recorded as filler score, and the wake-up score of target speech can be recorded as keyword score. The ultimate goal is to find the moment when the wake-up score of target speech is greater than the wake-up score of non-target speech, which is the wake-up speech moment. For example, the filler decoding scheme aims to improve the performance of the keyword detection system by explicitly modeling and filtering non-keywords. This strategy is particularly suitable for applications that require fast response and high accuracy. The keyword decoding scheme is mainly constructed by utilizing the Viterbi decoding principle. The main process is as follows: (1) If the path with the highest probability passes through a certain point in the fence network, then the subpath from the starting point to that point must also be the most probable path from the starting point to that point. (2) Assuming that there are k states at the i-th time, there are k shortest paths from the starting point to the k states at the i-th time, and the final shortest path must pass through one of them. (3) According to the above properties, when calculating the shortest path of the i+1 state, only the shortest path from the starting point to the current k state value and the shortest path from the current state value to the i+1 state value need to be considered. The keyword decoding scheme is also a process of accumulating the scores of each state, and finally selecting the node on the path with a cumulative score greater than the filler score. At this point, the candidate wake-up results of the frame-level wake-up model can be obtained, including the wake-up word, the wake-up score (in order to distinguish it from the subsequent wake-up scores, it can be called the first wake-up score here), and the start time and end time of the wake-up audio segment in the input audio.It should be noted that a higher awakening score indicates a higher possibility of actually needing awakening, and conversely, a lower awakening score indicates a lower possibility of actually needing awakening.

[0034] In a specific implementation scenario, as described above, the multiple wake-up models may include an end-to-end wake-up model, and the second a posteriori information output after the end-to-end wake-up model processes the input audio may be obtained, and the second a posteriori information may include: the second probability values ​​of the audio frames in the input audio respectively involving several wake-up words. Exemplarily, taking the specific number of the several wake-up words as M as an example, for the input audio containing N audio frames, the second a posteriori information may be expressed as an M*N probability matrix. In addition, the second a posteriori information may also include the second probability value that the audio frame in the input audio does not involve the wake-up word. In this case, the second a posteriori information may be expressed as a (M+1)*N probability matrix. On this basis, the candidate wake-up result of the end-to-end wake-up model may be obtained based on whether there are a continuous preset number of audio frames whose second probability value involving the same wake-up word is not less than the probability threshold. Exemplarily, for each wake-up word, it may be detected that the second probability values ​​of the wake-up word involving consecutive K (e.g., 2, 3, etc.) audio frames are not less than the probability threshold. If so, the wake-up word may be temporarily marked as wake-up, otherwise the wake-up word may be marked as not wake-up. After performing the above operations on each wake-up word, it can be detected whether there are multiple wake-up words marked as wake-up. If so, the average probability value can be calculated for each wake-up word, and the one with the larger average probability value can be decided based on this, and finally marked as wake-up. After marking the final wake-up word, the wake-up score (in order to distinguish it from the aforementioned wake-up score, it can be called the second wake-up score here) and the start time, end time, etc. of the wake-up audio segment in the input audio can be determined as the candidate wake-up result of the end-to-end wake-up model.

[0035] In a specific implementation scenario, still taking the multiple wake-up models including at least a frame-level wake-up model and an end-to-end wake-up model as an example, after obtaining the candidate wake-up result of the frame-level wake-up model and the candidate wake-up result of the end-to-end wake-up model, the target wake-up result (in this implementation scenario, the first wake-up result) can be obtained by combining the two. Specifically, in response to the result acquisition time of the end-to-end wake-up model being earlier than the result acquisition time of the frame-level wake-up model, the candidate wake-up result of the end-to-end wake-up model can be selected as the target wake-up result; in response to the same result acquisition time of the end-to-end wake-up model and the frame-level wake-up model, the candidate wake-up result of the end-to-end wake-up model can be selected as the target wake-up result; in response to the result acquisition time of the frame-level wake-up model being at least a second preset time length (such as 80ms, etc.) earlier than the result acquisition time of the end-to-end wake-up model, the candidate wake-up result of the frame-level wake-up model can be selected as the target wake-up result. In the above manner, the candidate wake-up results of different wake-up models are selected as the target wake-up result in different situations, and the target wake-up result can be adaptively determined according to the specific situation, which helps to further improve the self-applicability to different usage scenarios.

[0036] In one implementation scenario, as described above, at least one microphone (e.g., only one microphone for collecting echo signals) may be provided independently of the microphone for collecting audio at each seat to collect echo signals. On this basis, the second audio may be obtained based on the echo signal, and the wake-up engine may be called based on the second audio to generate a second wake-up instance, and the actual number of channels of the wake-up engine in the second wake-up instance is the number of channels of the second audio (e.g., when only one microphone for collecting echo signals is provided, the number of channels of the second audio is 1), so that the second audio may be input to the second wake-up instance and run to obtain the second wake-up result, and then it may be determined whether to output the first wake-up result as the voice wake-up result based on the first wake-up result and its first acquisition time and the second wake-up result and its second acquisition time. It should be noted that in actual application, relevant interfaces for retrieving wake-up results can be set for the first wake-up instance and the second wake-up instance respectively, and when the first wake-up instance and the second wake-up instance output the wake-up results respectively, the corresponding wake-up results can be obtained through the corresponding interfaces and the corresponding result acquisition time can be determined. In the above method, based on the first wake-up result, further reference is made to the second wake-up result analyzed based on the echo signal, which can minimize false wake-ups caused by self-excitation caused by signal reflection, and help improve the accuracy of voice wake-up. In addition, since the first wake-up instance and the second wake-up instance run independently, the efficiency of voice wake-up can also be improved from this perspective.

[0037] In a specific implementation scenario, the operation process of the second wake-up instance and the specific connotation of the second wake-up result can refer to the related description of the aforementioned first wake-up instance and the first wake-up result, which will not be repeated here.

[0038] In a specific implementation scenario, after obtaining the first wake-up result and the second wake-up result, in response to the first acquisition time of the first wake-up result being earlier than the second acquisition time of the second wake-up result, at least the second wake-up score in the second wake-up result can be acquired, and based on whether the second wake-up score exceeds the second wake-up threshold, it is determined whether to output the first wake-up result as the voice wake-up result. Exemplarily, in response to the second wake-up score not exceeding the second wake-up threshold, it can be determined to output the first wake-up result as the voice wake-up result, at which time the accuracy of the voice wake-up can be ensured as much as possible, whereas in response to the second wake-up score exceeding the second wake-up threshold, it can be determined to ignore the first wake-up result, at which time the possibility of false wake-up can be reduced as much as possible.

[0039] In a specific implementation scenario, after obtaining the first wake-up result and the second wake-up result, in response to the first acquisition time being later than the second acquisition time, a waiting detection result can be obtained based on whether the first acquisition time is within a first preset time length (e.g., 160ms) after the second acquisition time, and then based on the waiting detection result, it is determined whether to output the first wake-up result as a voice wake-up result. Exemplarily, in response to the waiting detection result including that the first acquisition time is within a first preset time length after the second acquisition time, the first wake-up result can be shielded, and the possibility of false wake-up can be reduced as much as possible. As a possible example, when the second wake-up result is obtained first, a timer of the first preset time length can be started, and it can be determined whether the first wake-up result is obtained again during this period. If so, the first wake-up result can be shielded to reduce the wake-up self-excitation as much as possible.

[0040] In an implementation scenario, please refer to Figure 2 , Figure 2 1 is a process diagram of an embodiment of the voice wake-up method of the present application. Figure 2As shown, as a possible example, the state detection results of each seat can be obtained by setting the seat sensor of each seat, and the original audio collected at the seat in the seated state represented by the state detection result can be further obtained to adaptively determine the number of target sound zones, and then the original audio can be preprocessed (such as echo cancellation processing, narrow beam algorithm processing, etc.) to obtain the first audio with the number of audio channels as the target number of sound zones. On this basis, the wake-up engine can be called based on the target number of sound zones to obtain the first wake-up instance, and the actual number of channels set by the wake-up engine in the first wake-up instance is automatically adjusted to the target number of sound zones, and then the first audio is used as the input parameter of the first wake-up instance and the first wake-up instance is run as a whole to perform the wake-up calculation to obtain the first wake-up result, so as to optimize the resource utilization efficiency and reduce unnecessary system loss. At the same time, in order to avoid wake-up self-excitation as much as possible, the wake-up engine can also be called based on the second audio obtained by the echo signal to obtain the second wake-up instance, and the actual number of channels of the wake-up engine in the second wake-up instance is the number of channels of the second audio (such as 1), and the final voice wake-up result is determined by combining the first wake-up result of running the first wake-up instance and the second wake-up result of running the second wake-up instance.

[0041] The above scheme obtains the audio collection results and state detection results at each seat, and the audio collection results include whether the audio is collected and the original audio when the audio is collected, and the state detection result package is whether the seat is in the sitting state, and then based on the audio collection results and the state detection results, the target number of sound zones is determined, and based on the state detection result, the first audio with the number of audio channels as the target number of sound zones is extracted from the audio collection results, so as to call the wake-up engine based on the target number of sound zones to obtain the first wake-up instance, and the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones, and then the first audio is input into the first wake-up instance and run to obtain the first wake-up result, so on the one hand, the first wake-up result obtained by calling the wake-up engine is run The wake-up instance can realize multi-zone wake-up. On the other hand, since the number of target zones is adaptively determined according to the audio acquisition results and the status detection results, and the actual number of channels of the wake-up engine in the first wake-up instance is the target zone number, compared with the target zone number set as a fixed zone number, the adaptability of multi-zone wake-up to different driving scenarios can be improved. On the other hand, since the number of audio channels of the first audio is also the target zone number adaptively determined, the interference audio of voice wake-up can be excluded as much as possible, and the wake-up accuracy of multi-zone wake-up for different driving scenarios can be improved as much as possible. Compared with processing the audio signals of all zones during the voice wake-up process, the resource consumption of multi-zone wake-up can be reduced as much as possible. Therefore, under the premise of realizing multi-zone wake-up, the wake-up accuracy and adaptability of multi-zone wake-up for different driving scenarios can be improved as much as possible, and the resource consumption of multi-zone wake-up can be reduced as much as possible.

[0042] See also Figure 3 , Figure 3 It is a schematic diagram of the framework of an embodiment of the voice wake-up device of the present application. The voice wake-up device 30 includes: a result acquisition module 31, a determination extraction module 32, a first calling module 33 and a first running module 34. The result acquisition module 31 is used to obtain the audio collection results and state detection results at each seat; wherein the audio collection results include whether the audio is collected and the original audio when the audio is collected, and the state detection results include whether it is in the sitting state; the determination extraction module 32 is used to determine the target number of sound zones based on the audio collection results and the state detection results, and based on the state detection results, extract the first audio whose audio channel number is the target number of sound zones from the audio collection results; the first calling module 33 is used to call the wake-up engine based on the target number of sound zones to obtain the first wake-up instance; wherein the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones; the first running module 34 is used to input the first audio to the first wake-up instance and run it to obtain the first wake-up result.

[0043] In the above scheme, the voice wake-up device 30 obtains the audio collection results and state detection results at each seat, and the audio collection results include whether the audio is collected and the original audio when the audio is collected, and the state detection result package is whether the seat is in the sitting state, and then based on the audio collection results and the state detection results, the target number of sound zones is determined, and based on the state detection result, the first audio with the number of audio channels as the target number of sound zones is extracted from the audio collection results, so as to call the wake-up engine based on the target number of sound zones to obtain the first wake-up instance, and the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones, and then the first audio is input into the first wake-up instance and run to obtain the first wake-up result, so on the one hand, the operation obtained by calling the wake-up engine The first wake-up instance obtained can realize multi-zone wake-up. On the other hand, since the number of target zones is adaptively determined according to the audio collection results and the status detection results, and the actual number of channels of the wake-up engine in the first wake-up instance is the target zone number, compared with the target zone number being set to a fixed zone number, the adaptability of multi-zone wake-up to different driving scenarios can be improved. On the other hand, since the number of audio channels of the first audio is also the target zone number adaptively determined, the interference audio of voice wake-up can be excluded as much as possible, and the wake-up accuracy of multi-zone wake-up for different driving scenarios can be improved as much as possible. Compared with processing the audio signals of all zones during the voice wake-up process, the resource consumption of multi-zone wake-up can be reduced as much as possible. Therefore, under the premise of realizing multi-zone wake-up, the wake-up accuracy and adaptability of multi-zone wake-up for different driving scenarios can be improved as much as possible, and the resource consumption of multi-zone wake-up can be reduced as much as possible.

[0044] In some disclosed embodiments, it is determined that the extraction module 32 includes a sound zone detection submodule, which is used to determine the sound zone detection result at the corresponding seat based on whether the audio collection result and the state detection result at the same seat are consistent in connotation; wherein the sound zone detection result includes whether to use the area where the seat is located as the sound zone; it is determined that the extraction module 32 includes a sound zone statistics submodule, which is used to perform statistics based on the sound zone detection results at each seat to obtain a target number of sound zones.

[0045] In some disclosed embodiments, the sound zone detection submodule includes a first determination unit for determining that the connotations are consistent and that the sound zone detection result at the corresponding seat includes taking the area where the corresponding seat is located as the sound zone in response to the audio collection result at the same seat including that audio is collected and the state detection result includes being in the seated state; the sound zone detection submodule includes a second determination unit for determining that the connotations are inconsistent and that the sound zone detection result at the corresponding seat includes not taking the area where the corresponding seat is located as the sound zone in response to the audio collection result at the same seat including that audio is collected and the state detection result includes not being in the seated state, or the audio collection result at the same seat including that audio is not collected and the state detection result includes being in the seated state.

[0046] In some disclosed embodiments, the voice wake-up device 30 includes a second calling module, which is used to obtain a second audio based on an echo signal, and to call a wake-up engine based on the second audio to generate a second wake-up instance; wherein the actual number of channels of the wake-up engine in the second wake-up instance is the number of channels of the second audio; the voice wake-up device 30 includes a second running module, which is used to input the second audio to the second wake-up instance and run it to obtain a second wake-up result; the voice wake-up device 30 includes a result decision module, which is used to determine whether to output the first wake-up result as a voice wake-up result based on the first wake-up result and its first acquisition time and the second wake-up result and its second acquisition time.

[0047] In some disclosed embodiments, the result decision module includes a first response submodule, which is used to obtain at least a second wake-up score in the second wake-up result in response to the first acquisition time being earlier than the second acquisition time; the result decision module includes an output determination submodule, which is used to determine whether to output the first wake-up result as a voice wake-up result based on whether the second wake-up score exceeds a second wake-up threshold.

[0048] In some disclosed embodiments, the output determination submodule includes a result output unit for determining to output the first wake-up result as a voice wake-up result in response to the second wake-up score not exceeding the second wake-up threshold; the output determination submodule includes a result ignoring unit for determining to ignore the first wake-up result in response to the second wake-up score exceeding the second wake-up threshold.

[0049] In some disclosed embodiments, the result decision module includes a second response submodule, which is used to obtain a waiting detection result in response to the first acquisition time being later than the second acquisition time, based on whether the first acquisition time is within a first preset time length after the second acquisition time; the output determination submodule is also used to determine whether to output the first wake-up result as a voice wake-up result based on the waiting detection result.

[0050] In some disclosed embodiments, the output determination submodule includes a result shielding unit for determining to shield the first wake-up result in response to the waiting detection result including that the first acquisition time is within a first preset time length after the second acquisition time.

[0051] In some disclosed embodiments, the wake-up engine is provided with a plurality of wake-up models, and the voice wake-up device 30 includes a candidate acquisition module for processing the input audio of the wake-up engine based on different wake-up models respectively, and obtaining the candidate wake-up results output by each wake-up model respectively; the voice wake-up device 30 includes a target determination module for analyzing based on each candidate wake-up result and its result acquisition time, and obtaining the target wake-up result finally output by the wake-up engine; wherein, when the input audio is the first audio, the target wake-up result is the first wake-up result, and when the input audio is the second audio representing the echo signal, the target wake-up result is the second wake-up result for the second audio.

[0052] In some disclosed embodiments, multiple wake-up models include at least a frame-level wake-up model and an end-to-end wake-up model, and the target determination module includes a first selection submodule for selecting a candidate wake-up result of the end-to-end wake-up model as the target wake-up result in response to the result acquisition time of the end-to-end wake-up model being earlier than the result acquisition time of the frame-level wake-up model; the target determination module includes a second selection submodule for selecting a candidate wake-up result of the end-to-end wake-up model as the target wake-up result in response to the result acquisition time of the end-to-end wake-up model and the frame-level wake-up model being the same; the target determination module includes a third selection submodule for selecting a candidate wake-up result of the frame-level wake-up model as the target wake-up result in response to the result acquisition time of the frame-level wake-up model being at least a second preset time length earlier than the result acquisition time of the end-to-end wake-up model.

[0053] In some disclosed embodiments, the candidate acquisition module includes a first a posteriori submodule, which is used to obtain the first a posteriori information output by the frame-level wake-up model after processing the input audio when multiple wake-up models include a frame-level wake-up model; wherein the first a posteriori information includes: first probability values ​​that the audio frames in the input audio belong to several pronunciation states respectively; the candidate acquisition module includes a first decoding submodule, which is used to decode the first a posteriori information based on a preset decoding strategy to obtain a candidate wake-up result of the frame-level wake-up model; wherein the preset decoding strategy at least includes filler-keyword.

[0054] In some disclosed embodiments, the candidate acquisition module includes a second a posteriori submodule, which is used to obtain second a posteriori information output by the end-to-end wake-up model after processing the input audio when multiple wake-up models include an end-to-end wake-up model; wherein the second a posteriori information includes: second probability values ​​of audio frames in the input audio respectively involving several wake-up words; the candidate acquisition module includes a second decoding submodule, which is used to obtain the candidate wake-up result of the end-to-end wake-up model based on whether there are a continuous preset number of audio frames whose second probability value involving the same wake-up word is not lower than a probability threshold.

[0055] See also Figure 4 , Figure 4 It is a schematic diagram of the framework of an embodiment of an electronic device of the present application. The electronic device 40 includes at least a memory 41 and a processor 42 coupled to each other, the memory 41 stores at least program instructions, and the processor 42 is used to execute the program instructions to implement the steps in any of the above-mentioned voice wake-up method embodiments. For details, please refer to the aforementioned disclosed embodiments, which will not be repeated here. As a possible example, the electronic device 40 may include but is not limited to a car computer, etc., and the specific type of the electronic device 40 is not limited here.

[0056] Specifically, the processor 42 is used to control itself and the memory 41 to implement the steps in any of the above-mentioned voice wake-up method embodiments. The processor 42 can also be called a CPU (Central Processing Unit). The processor 42 may be an integrated circuit chip with signal processing capabilities. The processor 42 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field-programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 42 can be implemented by an integrated circuit chip.

[0057] In the above scheme, the electronic device 40 obtains the audio collection results and state detection results at each seat, and the audio collection results include whether the audio is collected and the original audio when the audio is collected, and the state detection result package is whether the seat is in the sitting state, and then based on the audio collection results and the state detection results, the target number of sound zones is determined, and based on the state detection result, the first audio with the number of audio channels as the target number of sound zones is extracted from the audio collection results, so as to call the wake-up engine based on the target number of sound zones to obtain the first wake-up instance, and the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones, and then the first audio is input into the first wake-up instance and run to obtain the first wake-up result, so on the one hand, the operation obtained by calling the wake-up engine The first wake-up instance can realize multi-zone wake-up. On the other hand, since the number of target zones is adaptively determined according to the audio collection results and the status detection results, and the actual number of channels of the wake-up engine in the first wake-up instance is the target zone number, compared with the target zone number being set to a fixed zone number, the adaptability of multi-zone wake-up to different driving scenarios can be improved. On the other hand, since the number of audio channels of the first audio is also the target zone number adaptively determined, the interference audio of voice wake-up can be excluded as much as possible, and the wake-up accuracy of multi-zone wake-up for different driving scenarios can be improved as much as possible. Compared with processing the audio signals of all zones during the voice wake-up process, the resource consumption of multi-zone wake-up can be reduced as much as possible. Therefore, under the premise of realizing multi-zone wake-up, the wake-up accuracy and adaptability of multi-zone wake-up for different driving scenarios can be improved as much as possible, and the resource consumption of multi-zone wake-up can be reduced as much as possible.

[0058] See also Figure 5 , Figure 5 It is a schematic diagram of the framework of an embodiment of the voice wake-up system of the present application. The voice wake-up system 50 includes at least a vehicle computer 51, and a microphone 52 and a seat sensor 53 respectively arranged on each seat (not shown in the figure). The microphone 52 and the seat sensor 53 are respectively connected to the vehicle computer 51 for communication. The vehicle computer 51 includes at least the electronic equipment in the above-mentioned electronic equipment embodiment. For details, please refer to the aforementioned disclosed embodiment, which will not be repeated here. It should be noted that the microphone 52 is at least used to obtain the audio collection result at the seat, and the seat sensor 53 is used to obtain the status detection result at the seat. In addition, the voice wake-up system may also include a microphone 52 arranged at a position where audio reflection may exist, for collecting echo signals. For details, please refer to the relevant description in the aforementioned voice wake-up method embodiment, which will not be repeated here.

[0059] In the above scheme, the voice wake-up system 50 at least includes a vehicle computer 51, and a microphone 52 and a seat sensor 53 respectively arranged at each seat, the microphone 52 and the seat sensor 53 are respectively connected to the vehicle computer 51 for communication, the vehicle computer 51 includes at least the electronic device in the above electronic device embodiment, and the microphone 52 is at least used to obtain the audio collection result at the seat, and the seat sensor 53 is used to obtain the state detection result at the seat. Therefore, on the one hand, the first wake-up instance obtained by calling the wake-up engine can be run to realize multi-zone wake-up, on the other hand, since the target number of zones is adaptively determined according to the audio collection result and the state detection result, and the actual number of channels of the wake-up engine in the first wake-up instance is the target number of zones, compared with the target number of zones being set as a fixed number of zones, the adaptability of the multi-zone wake-up to different driving scenes can be improved, and on the other hand, since the number of audio channels of the first audio is also the target number of zones determined adaptively, the interference audio of the voice wake-up can be eliminated as much as possible, the wake-up accuracy of the multi-zone wake-up to different driving scenes can be improved as much as possible, and compared with processing the audio signals of all zones during the voice wake-up process, the resource consumption of the multi-zone wake-up can be reduced as much as possible. Therefore, under the premise of realizing multi-zone wake-up, the wake-up accuracy and adaptability of multi-zone wake-up to different driving scenarios can be improved as much as possible, and the resource consumption of multi-zone wake-up can be reduced as much as possible.

[0060] See also Figure 6 , Figure 6 The computer-readable storage medium 60 stores program instructions 61 that can be executed by a processor, and the program instructions 61 are used to implement the steps in any of the above-mentioned voice wake-up method embodiments.

[0061] In the above scheme, the computer-readable storage medium 60 obtains the audio collection results and state detection results at each seat, and the audio collection results include whether the audio is collected and the original audio when the audio is collected, and the state detection result package is whether the seat is in the sitting state, and then based on the audio collection results and the state detection results, the target number of sound zones is determined, and based on the state detection result, the first audio with the number of audio channels as the target number of sound zones is extracted from the audio collection results, so as to call the wake-up engine based on the target number of sound zones to obtain the first wake-up instance, and the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones, and then the first audio is input into the first wake-up instance and run to obtain the first wake-up result, so on the one hand, the operation performed by calling the wake-up engine The first wake-up instance obtained can realize multi-zone wake-up. On the other hand, since the number of target zones is adaptively determined according to the audio acquisition results and the status detection results, and the actual number of channels of the wake-up engine in the first wake-up instance is the target zone number, compared with the target zone number being set as a fixed zone number, the adaptability of multi-zone wake-up to different driving scenarios can be improved. On the other hand, since the number of audio channels of the first audio is also the target zone number adaptively determined, the interference audio of voice wake-up can be excluded as much as possible, and the wake-up accuracy of multi-zone wake-up for different driving scenarios can be improved as much as possible. Compared with processing the audio signals of all zones during the voice wake-up process, the resource consumption of multi-zone wake-up can be reduced as much as possible. Therefore, under the premise of realizing multi-zone wake-up, the wake-up accuracy and adaptability of multi-zone wake-up to different driving scenarios can be improved as much as possible, and the resource consumption of multi-zone wake-up can be reduced as much as possible.

[0062] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0063] The above description of various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other, and for the sake of brevity, they will not be repeated herein.

[0064] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0065] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0066] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0067] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0068] If the technical solution of this application involves personal information, the product using the technical solution of this application has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using the technical solution of this application has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, the personal information processing rules are notified by obvious signs / information, and the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A voice wake-up method, characterized in that: include: Acquire audio collection results and status detection results at each seat; wherein the audio collection results include whether audio is collected and the original audio when the audio is collected, and the status detection results include whether the seat is in an occupied state; Determine a target number of sound zones based on the audio collection result and the state detection result, and extract a first audio whose number of audio channels is the target number of sound zones from the audio collection result based on the state detection result; Based on the target number of sound zones, a wake-up engine is called to obtain a first wake-up instance; wherein the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones; The first audio is input to the first wake-up instance and executed to obtain a first wake-up result.

2. The method according to claim 1, characterized in that The determining the target number of sound zones based on the audio collection result and the state detection result includes: Determine the sound zone detection result at the corresponding seat based on whether the audio collection result and the state detection result at the same seat are consistent in connotation; wherein the sound zone detection result includes whether the area where the seat is located is used as the sound zone; Statistics are performed based on the sound zone detection results at each seat to obtain the target number of sound zones.

3. The method according to claim 2, characterized in that The determining of the sound zone detection result at the corresponding seat based on whether the audio collection result and the state detection result at the same seat are consistent in connotation includes at least one of the following: In response to the audio collection result at the same seat including audio being collected and the state detection result including being in the seated state, determining that the connotation is consistent and determining the sound zone detection result at the corresponding seat includes taking the area where the corresponding seat is located as the sound zone; In response to the audio collection result at the same seat including that audio is collected and the state detection result includes that it is not in the seated state, or the audio collection result at the same seat includes that audio is not collected and the state detection result includes that it is in the seated state, it is determined that the connotation is inconsistent and the sound zone detection result corresponding to the seat includes not taking the area where the corresponding seat is located as the sound zone.

4. The method according to claim 1, characterized in that: The method further comprises: Based on the echo signal, a second audio is obtained, and based on the second audio, the wake-up engine is called to generate a second wake-up instance; wherein the actual number of channels of the wake-up engine in the second wake-up instance is the number of channels of the second audio; Input the second audio to the second wake-up instance and run it to obtain a second wake-up result; Based on the first wake-up result and the first acquisition time thereof and the second wake-up result and the second acquisition time thereof, it is determined whether to output the first wake-up result as the voice wake-up result.

5. The method according to claim 4, characterized in that The determining, based on the first wake-up result and the first acquisition time thereof and the second wake-up result and the second acquisition time thereof, whether to output the first wake-up result as the voice wake-up result includes: In response to the first acquisition time being earlier than the second acquisition time, acquiring at least a second wake-up score in the second wake-up result; Based on whether the second wake-up score exceeds a second wake-up threshold, it is determined whether to output the first wake-up result as the voice wake-up result.

6. The method according to claim 5, characterized in that The determining whether to output the first wake-up result as the voice wake-up result based on whether the second wake-up score exceeds a second wake-up threshold includes at least one of the following: In response to the second wake-up score not exceeding the second wake-up threshold, determining to output the first wake-up result as the voice wake-up result; In response to the second wakeup score exceeding the second wakeup threshold, it is determined to ignore the first wakeup result.

7. The method according to claim 4, characterized in that The determining, based on the first wake-up result and the first acquisition time thereof and the second wake-up result and the second acquisition time thereof, whether to output the first wake-up result as the voice wake-up result includes: In response to the first acquisition time being later than the second acquisition time, obtaining a waiting detection result based on whether the first acquisition time is within a first preset time length after the second acquisition time; Based on the waiting detection result, it is determined whether to output the first wake-up result as the voice wake-up result.

8. The method according to claim 7, characterized in that The determining, based on the waiting detection result, whether to output the first wake-up result as the voice wake-up result at least includes: In response to the waiting detection result including that the first acquisition time is within a first preset time period after the second acquisition time, it is determined to shield the first wake-up result.

9. The method according to any one of claims 1 to 8, characterized in that: The wake-up engine is provided with a plurality of wake-up models, and the operation steps of the wake-up engine include: Processing the input audio of the wake-up engine based on different wake-up models respectively to obtain candidate wake-up results output by each wake-up model respectively; Analyze each of the candidate wake-up results and the time when the results are obtained to obtain the target wake-up result finally output by the wake-up engine; Among them, when the input audio is the first audio, the target wake-up result is the first wake-up result; when the input audio is the second audio representing the echo signal, the target wake-up result is the second wake-up result of the second audio.

10. The method according to claim 9, characterized in that The multiple wake-up models include at least a frame-level wake-up model and an end-to-end wake-up model, and the target wake-up result finally output by the wake-up engine is obtained by analyzing each of the candidate wake-up results and the time when the result is obtained, including at least one of the following: In response to the result acquisition time of the end-to-end wake-up model being earlier than the result acquisition time of the frame-level wake-up model, selecting the candidate wake-up result of the end-to-end wake-up model as the target wake-up result; In response to the same time for obtaining the results of the end-to-end wake-up model and the frame-level wake-up model, selecting a candidate wake-up result of the end-to-end wake-up model as the target wake-up result; In response to a result acquisition time of the frame-level wake-up model being at least a second preset time length earlier than a result acquisition time of the end-to-end wake-up model, a candidate wake-up result of the frame-level wake-up model is selected as the target wake-up result.

11. The method according to claim 9, characterized in that The multiple wake-up models include a frame-level wake-up model, and the input audio of the wake-up engine is processed based on different wake-up models to obtain candidate wake-up results output by each wake-up model, including: Obtaining first a posteriori information output after the frame-level wake-up model processes the input audio; wherein the first a posteriori information includes: first probability values ​​of audio frames in the input audio belonging to a plurality of pronunciation states; Decoding the first a posteriori information based on a preset decoding strategy to obtain a candidate wake-up result of the frame-level wake-up model; wherein the preset decoding strategy at least includes filler-keyword; And / or, the multiple wake-up models include an end-to-end wake-up model, and the processing of the input audio of the wake-up engine based on different wake-up models to obtain candidate wake-up results output by each wake-up model includes: Obtaining second a posteriori information output after the end-to-end wake-up model processes the input audio; wherein the second a posteriori information includes: second probability values ​​of audio frames in the input audio respectively involving a plurality of wake-up words; Based on whether there are a continuous preset number of audio frames involving the same wake-up word whose second probability value is not less than a probability threshold, a candidate wake-up result of the end-to-end wake-up model is obtained.

12. A voice wake-up device, characterized in that: include: A result acquisition module is used to acquire audio collection results and state detection results at each seat; wherein the audio collection results include whether audio is collected and the original audio when the audio is collected, and the state detection results include whether the seat is in an occupied state; Determine an extraction module, which is used to determine the target number of sound zones based on the audio collection result and the state detection result, and extract a first audio whose number of audio channels is the target number of sound zones from the audio collection result based on the state detection result; A first calling module is used to call the wake-up engine based on the target number of sound zones to obtain a first wake-up instance; wherein the actual number of channels of the wake-up engine in the first wake-up instance is the target number of sound zones; The first running module is used to input the first audio to the first wake-up instance and run it to obtain a first wake-up result.

13. An electronic device, characterized in that: The device at least comprises a memory and a processor coupled to each other, wherein the memory at least stores program instructions, and the processor is used to execute the program instructions to implement the voice wake-up method according to any one of claims 1 to 11.

14. A voice wake-up system, characterized in that: The invention comprises at least a vehicle computer, and microphones and seat sensors respectively arranged at each seat, wherein the microphones and seat sensors are respectively connected to the vehicle computer for communication, and the vehicle computer comprises at least the electronic device according to claim 13, wherein the microphone is at least used to obtain audio collection results at the seat, and the seat sensor is used to obtain status detection results at the seat.

15. A computer-readable storage medium, characterized in that: Program instructions that can be executed by a processor are stored, and the program instructions are used to implement the voice wake-up method according to any one of claims 1 to 11.

Images