Voice wake-up method and apparatus, electronic device, storage medium, and vehicle

Abstract

The present application relates to a kind of voice wake-up method, device, electronic equipment, storage medium and vehicle, by multiple microphone array, the audio data corresponding to multiple audio channels in vehicle respectively is continuously acquired, each audio channel corresponds to a microphone array;According to the audio data corresponding to multiple audio channels respectively, establish voiceprint feature database, and the reference voiceprint feature corresponding to each audio channel is included in voiceprint feature database;When detecting that there is wake-up instruction in the audio data of initial audio channel by target wake-up detection model, the real-time voiceprint feature of the audio data of initial audio channel is compared with voiceprint feature database, and voiceprint comparison result is obtained;According to voiceprint comparison result, determine target audio channel, and wake up target audio channel.

Description

Technical Field

[0001] This invention relates to the field of smart cockpit technology, specifically to a voice wake-up method, device, electronic device, storage medium, and vehicle. Background Technology

[0002] With the rapid iteration of smart cockpit technology, in-vehicle voice interaction has become the core interaction method for vehicle intelligence. Users can directly wake up the vehicle's intelligent system by speaking. To achieve full coverage of voice wake-up function for the driver, front passenger, and rear left and right seats, vehicles are generally equipped with a four-channel microphone array for full-domain voice signal acquisition. However, during the voice wake-up stage, the audio data collected by each channel microphone array is continuously monitored for wake-up, which greatly increases the computing load of the vehicle chip. Furthermore, due to the limited interior space of the vehicle, when the user speaks, multiple microphone arrays may collect the same user's wake-up command, which may result in some audio channels being ineffectively woken up. Summary of the Invention

[0003] One objective of this invention is to provide a voice wake-up method to solve the problems in the prior art where continuous wake-up monitoring is performed on the audio data collected by each channel microphone array, increasing the computational load of the vehicle chip; and where multiple microphone arrays can collect the same user's wake-up command, resulting in some audio channels being invalidally woken up. A second objective is to provide a voice wake-up device; a third objective is to provide an electronic device; a fourth objective is to provide a computer-readable storage medium; and a fifth objective is to provide a vehicle.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A voice wake-up method, the method comprising: The system continuously acquires audio data corresponding to multiple audio channels in the vehicle through a multi-channel microphone array, with each audio channel corresponding to one microphone array. A voiceprint feature database is established based on the audio data corresponding to the multiple audio channels, and the voiceprint feature database includes the baseline voiceprint features corresponding to each audio channel. When a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the real-time voiceprint features of the audio data of the initial audio channel are compared with the voiceprint feature database to obtain the voiceprint comparison result. Based on the voiceprint comparison results, the target audio channel is determined and the target audio channel is activated.

[0005] Based on the aforementioned technical methods, a wake-up detection model is used to filter wake-up commands for each audio channel, selecting the initial audio channel containing the wake-up command for subsequent voiceprint matching. This eliminates the need to perform voiceprint matching on the audio data of all audio channels, reducing computational load and improving computational efficiency. Furthermore, corresponding baseline voiceprint features are pre-stored for each audio channel. In the subsequent wake-up detection process, it is possible to accurately determine which specific audio channel and corresponding user triggered the wake-up command, thereby waking up the target audio channel corresponding to that user. Essentially, it is not about looking at which audio channel's audio data triggered the wake-up command, but directly pinpointing which user's speech triggered the wake-up command, thus improving the accuracy and anti-interference capability of voice wake-up.

[0006] A voice wake-up device, the device comprising: The acquisition module is used to continuously acquire audio data corresponding to multiple audio channels in the vehicle through a multi-microphone array, with each audio channel corresponding to one microphone array. The processing module is used to establish a voiceprint feature database based on the audio data corresponding to the multiple audio channels respectively. The voiceprint feature database includes the reference voiceprint features corresponding to each audio channel. The processing module is further configured to, when a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, compare the real-time voiceprint features of the audio data of the initial audio channel with the voiceprint feature database to obtain a voiceprint comparison result. The processing module is further configured to determine the target audio channel based on the voiceprint comparison result and to wake up the target audio channel.

[0007] An electronic device includes: a memory; a processor; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the voice wake-up method.

[0008] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the voice wake-up method.

[0009] A vehicle including the voice wake-up device.

[0010] The beneficial effects of this invention are: This invention uses a wake-up detection model to filter wake-up commands across various audio channels, selecting the initial audio channel containing the wake-up command for subsequent voiceprint matching. This eliminates the need to perform voiceprint matching on all audio channels, reducing computational load and improving efficiency. Furthermore, by pre-storing corresponding baseline voiceprint features for each audio channel, the invention can accurately determine which specific audio channel triggered the wake-up command during subsequent wake-up detection, thereby waking up the target audio channel corresponding to that user. Essentially, it doesn't rely on which audio channel's audio data triggered the wake-up command, but rather pinpoints exactly which user's speech triggered it, improving the accuracy and anti-interference capability of voice wake-up. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart of a voice wake-up method provided in an embodiment of the present invention; Figure 2 A flowchart of a voice wake-up method provided in another embodiment of the present invention; Figure 3 A flowchart of a voice wake-up method provided in another embodiment of the present invention; Figure 4 A flowchart of a voice wake-up method provided in another embodiment of the present invention; Figure 5 A flowchart of a voice wake-up method provided in another embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of the voice wake-up device provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0014] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0015] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0016] With the rapid iteration of intelligent cockpit technology, in-vehicle voice interaction has become the core interaction method for vehicle intelligence. In order to achieve multi-zone voice wake-up function covering the driver, front passenger and rear left and right seats, existing intelligent in-vehicle systems are generally equipped with a four-channel microphone array to collect voice signals in the whole vehicle. There are several implementation schemes for multi-channel voice wake-up methods.

[0017] Option 1 employs a full-channel parallel wake-up detection mechanism, which independently executes complete voice wake-up model inference calculations on the voice data collected from the four microphones to achieve full coverage wake-up recognition across multiple voice zones. However, it's conceivable that executing complete wake-up model inferences synchronously and in parallel on the four channels of voice data would require the same voice calculation operation to be performed four times, multiplying the computational load. Given the limited computing power and power consumption resources of the in-vehicle embedded chip, full-channel parallel computing would lead to excessively high system computing power occupancy and a surge in power consumption, easily causing voice response delays and system stuttering, failing to meet the low-power real-time interaction requirements of in-vehicle systems. Furthermore, the lack of effective channel filtering and noise reduction mechanisms means that not all channels will capture valid human voices; some channels may contain a large amount of invalid audio with no voice or pure noise that still participates in the entire inference process, resulting in significant redundant invalid calculations and further exacerbating computing power consumption.

[0018] Option 2 addresses the issue of numerous invalid channels (those with no speech and only noise) consuming computational resources in the four-channel configuration. A layered speech processing architecture with two-stage separation and two-stage wake-up can be employed. The first stage uses a small model to filter a large amount of negative sample data, saving computational resources, while the second stage uses confirmation to improve the wake-up rate. Specifically, a neural network model sequentially performs streaming pre-wake-up and offline secondary verification. Sound source separation combined with a multi-stage wake-up mechanism improves wake-up accuracy in complex scenarios. Furthermore, multi-task design integrates sound source localization and voice extraction capabilities to optimize recognition performance. However, the overall architecture still relies on multi-level inference based on the entire audio channel without pre-channel screening. Invalid noise channels continuously participate in the entire process of two-level separation and two-level wake-up, and the additional computational overhead brought by the multi-level model further amplifies the computing power and power consumption pressure on the automotive chip. It is still easy to produce response latency on low-computing-power automotive terminals. In addition, the second level of the two-level architecture adopts secondary offline voice separation + four-way offline wake-up, which does not significantly reduce computational complexity and is not conducive to the lightweight deployment of automotive embedded devices. This solution only performs layered optimization in the audio dimension and does not utilize multi-channel voiceprint and sound field correlation features to carry out cross-channel cross-verification. In the face of strong interference scenarios such as vehicle engine noise, wind noise, cabin reverberation, and multiple people talking at the same time, there are still cases of channel misjudgment and wake-up missed / false detection.

[0019] Option 3 addresses the issue of which audio region is activated during multi-region wake-up by introducing multimodal information. This option relies on in-vehicle cameras for personnel verification and position correction. However, under conditions such as camera obstruction, low light at night, passengers looking down / wearing masks, strong interior glare, and obstructed rear-seat visibility, it fails to effectively extract facial and mouth features, leading to visual positioning failure, the collapse of the dual-modal verification mechanism, and ultimately, degradation to pure audio positioning, resulting in the recurrence of audio region misjudgment. Furthermore, this option requires accompanying in-vehicle visual acquisition hardware, increasing the overall vehicle hardware modification cost and deployment threshold.

[0020] In summary, current solutions for audio channel wake-up generally suffer from the following problems: (1) High computational redundancy and excessive power consumption: The traditional four-channel wake-up adopts a full-channel parallel complete model inference mechanism. The four repeated wake-up calculations result in double the computing power consumption, highlighting the computing power bottleneck of the vehicle embedded hardware, which cannot meet the requirements of low power consumption and low latency operation.

[0021] (2) High proportion of invalid calculations: There is no pre-speech channel filtering mechanism, and the noise channel and empty channel continue to participate in wake-up inference. A large amount of invalid calculations occupy system resources, further reducing the real-time performance of voice interaction.

[0022] (3) Poor wake-up accuracy and stability: Relying solely on model inference to determine the wake-up channel is susceptible to interference from complex vehicle noise and reverberation, resulting in channel prediction errors, false wake-up triggers, and missed wake-up of target users. The multi-level model has insufficient adaptability.

[0023] The following is about Figure 1 The voice wake-up method shown below will be introduced, and the specific steps of this method are as follows: 101. Continuously acquire audio data corresponding to multiple audio channels in the vehicle through a multi-channel microphone array.

[0024] Each audio channel corresponds to one microphone array.

[0025] In this embodiment, a multi-channel microphone array can be installed inside the vehicle. Typically, for a typical five-seater sedan, four microphone arrays are installed, corresponding to the driver's seat, passenger seat, left rear seat, and right rear seat, respectively, to collect audio data from the corresponding positions. After the vehicle is started, the multi-channel microphone array is activated simultaneously, continuously collecting audio data corresponding to multiple audio channels. Here, there is a one-to-one correspondence between the audio channels and the microphone arrays, meaning that each microphone array has one audio channel, and the audio data corresponding to that audio channel is obtained through the corresponding microphone array. This audio data mainly includes the voice of the user at the position corresponding to the microphone array. Of course, since the interior of the vehicle can be a closed environment, the microphone array may also collect the voices of users in other positions, as well as environmental sounds inside the vehicle, such as ambient noise, wind noise, engine noise, and cabin echo.

[0026] 102. Establish a voiceprint feature database based on the audio data corresponding to multiple audio channels.

[0027] In this embodiment of the application, after obtaining the audio data corresponding to multiple audio channels, a voiceprint feature database can be established. The voiceprint feature database can store the reference voiceprint features corresponding to each audio channel. The reference voiceprint features are the voiceprints of the users located at the corresponding positions of the audio channels. For example, for the driver's audio channel, the voice of the user sitting in the driver's seat can be extracted from the audio data of the driver's audio channel, and then the voiceprint features of the user's voice can be extracted to construct the reference voiceprint features of the driver's audio channel. For each audio channel, the reference voiceprint features of the corresponding user are extracted, thereby constructing the voiceprint feature database.

[0028] 103. When a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the real-time voiceprint features of the audio data of the initial audio channel are compared with the voiceprint feature database to obtain the voiceprint comparison result.

[0029] In this embodiment, during the continuous acquisition of audio data by the microphone array, it is necessary to continuously determine whether the user's speech triggers a wake-up command. Therefore, the target wake-up detection model can monitor the audio data of each audio channel. If the target wake-up detection model detects a wake-up command in the audio data of the initial audio channel, it indicates that a user has spoken at the initial audio channel and triggered the wake-up command. However, it is conceivable that the vehicle space is limited, and sound may be reflected or crosstalk may occur inside the vehicle. When the microphone array at a certain position is acquiring audio data, it may simultaneously acquire the voices of users at multiple positions. Therefore, if the audio channel to be woken up is determined solely by identifying the wake-up command in the audio data, it is easy to cause the audio channel to be woken up incorrectly.

[0030] For example, if a passenger in the front passenger seat leans towards the driver and triggers a wake-up command while speaking, the microphone array on the driver's side picks up the passenger's voice and recognizes the wake-up command. At this point, the audio channel on the driver's side will be activated, even though it is actually the passenger speaking, resulting in an incorrect audio channel activation. Alternatively, while the passenger is speaking, the microphone array on the passenger's side will detect the wake-up command triggered by the passenger's speech. Simultaneously, the microphone arrays in the driver's seat and the rear seats will also pick up the passenger's voice. In this case, the audio data collected by each audio channel is equivalent to containing a wake-up command, causing all audio channels to be activated repeatedly, resulting in a chaotic audio channel configuration.

[0031] To avoid the aforementioned anomalies, this application can further compare the warming characteristics of the audio data upon detecting a wake-up command. This allows determination of which user's speech triggered the wake-up command. Therefore, the real-time voiceprint features of the initial audio channel's audio data can be compared with the voiceprint feature database to obtain a voiceprint comparison result. Since the voiceprint feature database stores the baseline voiceprint features corresponding to each audio channel, the comparison of voiceprint features can determine which user's speech triggered the wake-up command for that audio channel, thus enabling the subsequent wake-up of the corresponding audio channel. This voiceprint comparison result can then be used to indicate that the real-time voiceprint features of the initial audio channel's audio data are consistent with the baseline voiceprint features corresponding to that specific audio channel in the voiceprint feature database.

[0032] In some embodiments, the wake-up command can be a pre-set wake-up method, which is typically a wake-up word mode. That is, some fixed wake-up words are set in advance, and as long as the user's speech includes any wake-up word, it is considered that the user has triggered the wake-up command. Of course, other wake-up methods can also be used, such as detecting the user's wake-up intention or detecting the user's specific command content for wake-up, etc. This application does not make specific limitations.

[0033] 104. Based on the voiceprint comparison results, determine the target audio channel and wake up the target audio channel.

[0034] In this embodiment of the application, based on the voiceprint comparison results, it can be determined which user the current real-time voiceprint feature belongs to, thereby determining the target audio channel. The target audio channel is the audio channel corresponding to the baseline voiceprint feature that matches the real-time voiceprint feature in the voiceprint feature database, and then the target audio channel can be activated.

[0035] In this embodiment, a wake-up detection model is used to filter wake-up commands for each audio channel, selecting the initial audio channel containing the wake-up command for subsequent voiceprint matching. This eliminates the need to perform voiceprint matching on the audio data of all audio channels, reducing computational load and improving computational efficiency. Furthermore, corresponding baseline voiceprint features are pre-stored for each audio channel. In the subsequent wake-up detection process, it is possible to accurately determine which audio channel and corresponding user triggered the wake-up command, thereby waking up the target audio channel corresponding to that user. This is equivalent to directly pinpointing which user's speech triggered the wake-up command, rather than looking at which audio channel's audio data triggered the wake-up command, thus improving the accuracy and anti-interference capability of voice wake-up.

[0036] Figure 2 A flowchart of a voice wake-up method provided in another embodiment of the present invention is shown below. Figure 2 As shown, the method includes the following steps: 201. Collect raw mixed voice data from the vehicle using a multi-microphone array.

[0037] In this embodiment of the application, although the multi-microphone array set in the vehicle is set in multiple locations, the actual audio data collected is still mixed together. Therefore, the original mixed voice data collected by the multi-microphone array can be obtained. The original mixed voice data includes the audio data collected by each microphone array, which may specifically include complex human voices, environmental noise, wind noise, tire noise, music, conversation noise, engine noise, cabin echo, etc.

[0038] 202. Using an audio decoupling algorithm, the original mixed speech data is separated into multiple audio channels to obtain audio data corresponding to each channel.

[0039] In this embodiment, since the original mixed speech data is audio data collected by multiple microphone arrays, and due to the confined space inside the vehicle and the ease with which sound is reflected and crosstalk occurs, the mixed speech signals will have serious crosstalk problems. Therefore, in order to process each audio data separately and accurately filter out pure noise channels, the original mixed speech data can be audio separated, which is equivalent to separating the audio data collected by each microphone array. Specifically, an audio decoupling algorithm can be used to remove the cross-sound interference between each audio channel and output multiple clean, non-interfering single-channel speech signals, thereby obtaining audio data corresponding to multiple audio channels respectively.

[0040] It should be noted that this audio decoupling algorithm is used to completely separate the different sound source features that originally interfered with each other from multi-channel mixed speech signals, obtaining a single-channel clean speech that does not interfere with each other, thus solving the problems of sound reflection and channel crosstalk in the vehicle. By using features orthogonalization and nonlinear mapping, the different sound source information in the mixed audio is decoupled, avoiding the overlap of sound features from different channels. Generally, relative gain arrays and independent component analysis algorithms can be used to achieve audio separation; alternatively, dual-hidden-layer backpropagation neural network models and dual-branch network models can be used for speech feature decoupling. This application does not specifically limit the implementation of this algorithm.

[0041] 203. Establish a voiceprint feature database based on the audio data corresponding to multiple audio channels.

[0042] 204. When a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the real-time voiceprint features of the audio data of the initial audio channel are compared with the voiceprint feature database to obtain the voiceprint comparison result.

[0043] 205. Based on the voiceprint comparison results, determine the target audio channel and wake up the target audio channel.

[0044] In this embodiment, the description of steps 203 to 205 is the same as the detailed description of steps 102 to 104 in the above embodiments, and will not be repeated in this embodiment.

[0045] Figure 3 A flowchart of a voice wake-up method provided in another embodiment of the present invention is shown below. Figure 3 As shown, the method includes the following steps: 301. Continuously acquire audio data corresponding to multiple audio channels in the vehicle through a multi-channel microphone array.

[0046] In this embodiment, the description of step 301 is the same as the detailed description of step 101 in the above embodiments, and will not be repeated in this embodiment.

[0047] 302. When a human voice signal is detected in the audio data corresponding to the audio channel, the audio channel is determined to be a valid channel.

[0048] In this embodiment, after obtaining the audio data corresponding to each audio channel, the audio data of each audio channel can be processed separately. This processing can include activity detection, that is, determining whether the audio data of this audio channel includes human voice. Since the audio data of each audio channel has been separated, the human voice included in the audio data of each audio channel is the voice of the user at the corresponding position of this audio channel. For example, if the audio data of the audio channel for the driver's seat includes human voice, then this voice is definitely the voice of the driver; if the audio data of the audio channel for the rear right side includes human voice, then this voice is definitely the voice of the user sitting on the rear right side. Therefore, human voice detection is performed on the audio data corresponding to each audio channel, and the audio channel with human voice signal is determined as a valid channel; while for the audio channel where no human voice signal is detected, its audio data may only contain ambient noise inside the car, indicating that the user sitting on that side of the audio channel is not speaking, or there is no user on that side of the audio channel, it is an empty seat, and no further processing is performed.

[0049] 303. Extract voiceprint features from the voice signals of the effective channels to obtain the baseline voiceprint features corresponding to the effective channels.

[0050] In this embodiment of the application, after determining the effective channel where there is a human voice signal, voiceprint features can be extracted from the human voice signal. Here, the human voice signal can be considered as the voice of the user at the corresponding position of the audio channel, and can be considered as the reference sound of the audio channel. Therefore, the voiceprint features extracted from the human voice signal can be used as the reference voiceprint features corresponding to the effective channel.

[0051] 304. Summarize the baseline voiceprint features of all valid channels and establish a voiceprint feature database.

[0052] In this embodiment of the application, after extracting the reference voiceprint features for each valid channel, the correspondence between the reference voiceprint features and the audio channels can be established, thereby constructing a voiceprint feature database.

[0053] In some embodiments, for the human voice signal of each effective channel, core voiceprint parameters such as human voice spectrum, fundamental frequency, and formant are screened, and the extracted channel-specific voiceprint parameters are encrypted, solidified, and stored to establish a benchmark feature library that matches channels and voiceprints one by one, thus completing the voiceprint registration for each channel.

[0054] In some embodiments, since there may be no user in some positions inside the vehicle, or the user in some positions may not be speaking, not all audio channels are valid channels. Therefore, it can be assumed that the process of building a voiceprint feature database is continuous until all audio channels are valid channels, that is, human voice signals are detected in the audio data of all audio channels.

[0055] For example, suppose the vehicle interior has microphone arrays installed in four areas: the driver's seat, the front passenger seat, the left rear seat, and the right rear seat. Currently, only the driver's seat, the front passenger seat, and the left rear seat have users. When the microphone arrays collect audio data, only the driver in the driver's seat is speaking. Therefore, only the driver's seat audio channel contains a human voice signal, making it a valid channel. The voiceprint feature database established after voiceprint feature extraction only includes the baseline voiceprint features of the driver's seat user. The other three audio channels are invalid channels and do not contain baseline voiceprint features. However, the audio data inside the vehicle is continuously monitored. As long as the front passenger's voice is collected by the microphone array in the front passenger seat at any moment, the baseline voiceprint features of that front passenger can be extracted, thus updating the voiceprint feature database. Similarly, for the left rear seat, as long as the rear left seat's voice is collected by the microphone array in the rear left seat at any moment, the baseline voiceprint features of that rear left seat user can be extracted, thus updating the voiceprint feature database.

[0056] In some embodiments, after the voiceprint feature database is established, the baseline voiceprint features corresponding to each audio channel stored therein do not remain unchanged. In real-world scenarios, the passenger in the front seat may move to the back seat, and the passenger in the back seat may move freely to the left and right. The driver in the driver's seat may also change. In this case, the baseline voiceprint features in the established voiceprint feature database may no longer accurately correspond to the audio channels. Therefore, the baseline voiceprint features corresponding to each audio channel stored in the voiceprint feature database can be periodically verified. This process can also be performed synchronously when a human voice signal is detected in the vehicle. For example, if the audio data of a certain audio channel includes a human voice signal, and the baseline voiceprint features of that audio channel are already stored in the voiceprint feature database, then the voiceprint features of that human voice signal can be extracted again and compared with the baseline voiceprint features already stored in the voiceprint feature database to see if they belong to the same user. If the comparison result shows that the voiceprint features are consistent, then no processing is performed; however, if the comparison result shows that the voiceprint features are inconsistent, it means that the user corresponding to this audio channel has changed, and then the baseline voiceprint features of that audio channel already stored in the voiceprint feature database can be updated.

[0057] In this embodiment, relying on the multi-channel dedicated voiceprint registration and comparison error correction mechanism, the problem of multi-channel false wake-up in complex vehicle noise and reverberation environments is effectively solved, and the accuracy and anti-interference ability of multi-zone voice wake-up in the whole vehicle are greatly improved.

[0058] 305. When a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the real-time voiceprint features of the audio data of the initial audio channel are compared with the voiceprint feature database to obtain the voiceprint comparison result.

[0059] 306. Based on the voiceprint comparison results, determine the target audio channel and wake up the target audio channel.

[0060] In the embodiments of this application, the description of steps 305 to 306 is the same as the detailed description of steps 103 to 104 in the above embodiments, and will not be repeated in the embodiments of this application.

[0061] Figure 4 A flowchart of a voice wake-up method provided in another embodiment of the present invention is shown below. Figure 4 As shown, the method includes the following steps: 401. Continuously acquire audio data corresponding to multiple audio channels in the vehicle through a multi-channel microphone array.

[0062] 402. Establish a voiceprint feature database based on the audio data corresponding to multiple audio channels.

[0063] In this embodiment, the description of steps 401 to 402 is the same as the detailed description of steps 101 to 102 in the above embodiments, and will not be repeated in this embodiment.

[0064] 403. Filter the audio data from multiple audio channels to determine the initial audio channel.

[0065] In this embodiment of the application, during the continuous collection of in-vehicle audio data, it is also necessary to continuously determine whether the user's speech has triggered a wake-up command. Since each microphone array is collecting audio data, but not every audio channel has someone speaking, the audio data of multiple audio channels can be filtered first to determine the initial audio channel. The audio data of the initial audio channel contains valid human voices. The number of initial audio channels is not limited. As long as the audio data includes valid human voices, it can be considered as the initial audio channel and enter the subsequent calculation.

[0066] It should be noted that the interior of an actual vehicle contains various sounds such as ambient noise, wind noise, tire noise, music, conversation noise, engine noise, and cabin echo. The algorithm can filter out invalid channels with no human voices and only retain the initial audio channels containing valid human voice signals for subsequent calculations.

[0067] 404. Perform content recognition on the audio data of the initial audio channel to obtain the audio content corresponding to the initial audio channel.

[0068] In this embodiment of the application, after extracting the initial audio channel containing human voice signals, it is necessary to detect whether a wake-up command has been triggered. Therefore, it is necessary to analyze what the user said. Content recognition can be performed on the audio data of the initial audio channel to determine the audio content corresponding to the initial audio channel. This can be achieved using common audio recognition algorithms, which convert the audio into text content.

[0069] 405. Using a target wake-up detection model, the audio content corresponding to the initial audio channel is detected. When a preset wake-up content is detected in the audio content corresponding to the initial audio channel, it is determined that a wake-up command exists in the audio data of the initial audio channel.

[0070] In this embodiment of the application, after obtaining the audio content corresponding to the initial audio channel, the target wake-up detection model can be used to detect the audio content corresponding to the initial audio channel and determine whether there is a preset wake-up content in the audio content. The preset wake-up content can be a pre-set wake-up word. If the target wake-up detection model detects that there is a preset wake-up content in the audio content corresponding to the initial audio channel, it means that the user at the corresponding position of the initial audio channel has triggered a wake-up command, that is, there is a wake-up command in the audio data of the initial audio channel.

[0071] In some embodiments, when the target wake-up detection model detects the audio content corresponding to the initial audio channel, it can perform hierarchical detection using a two-level wake-up model. Specifically, the target wake-up detection model detects the audio content corresponding to the initial audio channel. When a preset wake-up content is detected in the audio content corresponding to the initial audio channel, it is determined that a wake-up command exists in the audio data of the initial audio channel. This can be further categorized as follows: detecting the audio content corresponding to the initial audio channel using a first wake-up detection model to obtain a first detection result; when the first detection result indicates that a preset wake-up content exists in the audio content corresponding to the initial audio channel, detecting the audio content corresponding to the initial audio channel again using a second wake-up detection model to obtain a second detection result; when the second detection result indicates that a preset wake-up content still exists in the audio content corresponding to the initial audio channel, it is determined that a wake-up command exists in the audio data of the initial audio channel, where the model parameters of the first wake-up detection model are smaller than those of the second wake-up detection model.

[0072] It should be noted that both the first and second wake-up detection models are used to detect the presence of preset wake-up content in audio content. However, their detection accuracy differs. The first wake-up detection model can be considered a lightweight coarse-screening model. It has been specifically optimized and features fast inference speed, low resource consumption, and high wake-up recall. Its design allows for a small number of false positives, prioritizing ensuring no missed wake-up detections. The first wake-up detection model performs preliminary wake-up feature detection on each initial audio channel, obtaining a first detection result. This first detection result indicates whether preset wake-up content exists in the audio content of the initial audio channel, filtering out candidate channels with wake-up word features and directly eliminating human voice channels without wake-up intent.

[0073] Then, the second wake-up detection model can be a high-precision wake-up model. This second wake-up detection model only performs a secondary detection on the initial audio channels that have already passed the screening by the first wake-up detection model. This can be considered a verification process, accurately identifying the speech content, determining whether it is a legitimate wake-up command, and obtaining a second detection result. This second detection result can be considered as further indicating, based on the first detection result, whether the preset wake-up content actually exists in the audio content of the initial audio channels that the first detection result considers to contain it. If the second wake-up detection model considers the preset wake-up content to exist in the audio content of the initial audio channels, it means that the audio content of the initial audio channels has undergone two levels of model detection with different precision. Therefore, it can be considered that the preset wake-up content does indeed exist in the audio content of the initial audio channels, and the corresponding audio channel can be woken up. In other words, the first wake-up detection model first preliminarily screens the audio content of these initial audio channels to see if preset wake-up content exists. If it considers that the audio content of an initial audio channel might contain preset wake-up content, it sends it to the second wake-up detection model for further detection. The second wake-up detection model will then accurately detect which initial audio channel's audio content does indeed contain preset wake-up content.

[0074] In some embodiments, both the first wake-up detection model and the second wake-up detection model are pre-trained models. During training, a large amount of historical wake-up sample data can be collected, and the model parameters are continuously optimized. However, since the first wake-up detection model and the second wake-up detection model can be set as two models with different levels of accuracy, the first wake-up detection model allows for a certain degree of false recognition, just to ensure that no false detections are missed; while the second wake-up detection model needs to determine very accurately whether there is preset wake-up content. Therefore, during training, the convergence condition of the second wake-up detection model will be more stringent than that of the first wake-up detection model, and the model parameters of the first wake-up detection model are smaller than those of the second wake-up detection model.

[0075] In this embodiment, filtering via the pre-voice channel completely eliminates invalid inference from noisy, empty channels. Layered filtering through cascaded large and small models avoids inference from a four-channel, fully parallel, high-precision model, optimizing the traditional four full-scale, high-computational calculations into precise calculations across a small number of channels. This perfectly adapts to low-computational-power, low-power hardware scenarios in automotive embedded systems, completely resolving the issues of large computational loads and system lag / latency associated with multi-channel parallel computation. Furthermore, the first-level lightweight model ensures full coverage across multiple audio zones and eliminates missed wake-ups, resolving the problem of missed wake-up detection in multiple automotive regions. The second-level high-precision model undergoes secondary verification, combined with voiceprint feature error correction, effectively filtering false wake-up signals and addressing the technical challenge of traditional single-level models simultaneously achieving high recall and low false recognition.

[0076] 406. When a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the real-time voiceprint features of the audio data of the initial audio channel are compared with the voiceprint feature database to obtain the voiceprint comparison result.

[0077] 407. Based on the voiceprint comparison results, determine the target audio channel and wake up the target audio channel.

[0078] In this embodiment, the description of steps 406 to 407 is the same as the detailed description of steps 103 to 104 in the above embodiments, and will not be repeated in this embodiment.

[0079] Figure 5 A flowchart of a voice wake-up method provided in another embodiment of the present invention is shown below. Figure 5 As shown, the method includes the following steps: 501. Continuously acquire audio data corresponding to multiple audio channels in the vehicle through a multi-channel microphone array.

[0080] 502. Establish a voiceprint feature database based on the audio data corresponding to multiple audio channels.

[0081] In this embodiment, the description of steps 501 to 502 is the same as the detailed description of steps 101 to 102 in the above embodiments, and will not be repeated in this embodiment.

[0082] 503. When a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the baseline voiceprint features corresponding to the initial audio channel are extracted from the voiceprint feature database.

[0083] In this embodiment of the application, when a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, it is necessary to further confirm whether the audio data of the initial audio channel is the audio data issued by the user at the corresponding position of the initial audio channel. Here, the reference voiceprint features stored in the previously created voiceprint feature database can be used. Therefore, the reference voiceprint features corresponding to the initial audio channel can be extracted from the voiceprint feature database. The reference voiceprint features corresponding to the initial audio channel are the voiceprint features of the user at the corresponding position of the initial audio channel when the voiceprint feature database was created.

[0084] 504. Compare the real-time voiceprint features of the audio data of the initial audio channel with the baseline voiceprint features corresponding to the initial audio channel to obtain the first voiceprint comparison result.

[0085] In this embodiment of the application, in order to determine whether the audio data of the initial audio channel is the audio data emitted by the user at the corresponding location, the real-time voiceprint features of the audio data of the initial audio channel can be extracted. Then, the voiceprint features are compared with the baseline voiceprint features corresponding to the initial audio channel stored in the voiceprint feature database to obtain a first voiceprint comparison result. The first voiceprint comparison result can include comparison pass or comparison fail. If the comparison pass, it means that the audio data of the current initial audio channel is the sound emitted by the user at the corresponding location, and the voiceprint features match. If the comparison fails, it means that the audio data of the current initial audio channel is not the sound emitted by the user at the corresponding location, and the voiceprint features do not match.

[0086] In some embodiments, the process of comparing voiceprint features can be specifically determined by comparing the similarity between two voiceprint features. Specifically, the real-time voiceprint features of the audio data of the initial audio channel are compared with the reference voiceprint features corresponding to the initial audio channel to obtain a first voiceprint comparison result. This can specifically include: comparing the real-time voiceprint features of the audio data of the initial audio channel with the reference voiceprint features corresponding to the initial audio channel to determine the first voiceprint similarity between the real-time voiceprint features and the reference voiceprint features corresponding to the initial audio channel; when the first voiceprint similarity is detected to be greater than or equal to a preset similarity threshold, the first voiceprint comparison result is determined to be a successful comparison; when the first voiceprint similarity is detected to be less than the preset similarity threshold, the first voiceprint comparison result is determined to be a failed comparison.

[0087] It should be noted that the voiceprint characteristics of the same user's voice are basically unchanged. Therefore, if the audio data of the current initial audio channel is the voice of the user at the corresponding location, the first voiceprint similarity between the real-time voiceprint characteristics and the baseline voiceprint characteristics of the initial audio channel should be relatively high. Thus, a preset similarity threshold can be set. If the first voiceprint similarity is greater than or equal to the preset similarity threshold, then the audio data of the current initial audio channel can be considered to be the voice of the user at the corresponding location, and the first voiceprint comparison result is a pass. Conversely, if the first voiceprint similarity is less than the preset similarity threshold, then the audio data of the current initial audio channel can be considered not to be the voice of the user at the corresponding location, and the first voiceprint comparison result is a fail.

[0088] 505. If the first voiceprint comparison result indicates that the comparison fails, the real-time voiceprint features of the audio data of the initial audio channel are compared with the baseline voiceprint features of other audio channels in the voiceprint feature database to obtain the second voiceprint comparison result.

[0089] In this embodiment, if the first voiceprint comparison result indicates a failure, it means that the user currently sending the audio data of the initial audio channel is not the user who was previously located at the initial audio channel position. In order to accurately wake up the corresponding user's audio channel, it is possible to determine whether the real-time voiceprint features in the voiceprint feature database are consistent with the reference voiceprint features corresponding to the audio channels at other positions, and whether the audio data is sent by a user at another position. Therefore, the real-time voiceprint features of the audio data of the initial audio channel can be compared with the reference voiceprint features corresponding to other audio channels in the voiceprint feature database to obtain a second voiceprint comparison result. The second voiceprint comparison result can also include a successful comparison or a failed comparison. A successful comparison means that the audio data of the current initial audio channel is the sound sent by a user at another position in the voiceprint feature database, and it matches the reference voiceprint features of another audio channel. A failed comparison means that the audio data of the current initial audio channel is not the sound sent by a user at any other position, and it does not match the reference voiceprint features of any other audio channel.

[0090] In some embodiments, the process of comparing the real-time voiceprint features of the audio data of the initial audio channel with the baseline voiceprint features of other audio channels in the voiceprint feature database is also to calculate the second voiceprint similarity between the real-time voiceprint features and the baseline voiceprint features of other audio channels in the voiceprint feature database, and then compare it with a preset similarity threshold to obtain the comparison result. You can refer to the description of step 504, which will not be repeated here.

[0091] 506. If the first voiceprint comparison result indicates that the comparison is successful, the initial audio channel is determined as the target audio channel.

[0092] In this embodiment of the application, if the first voiceprint comparison result indicates that the comparison has passed, it means that the audio data of the initial audio channel is the sound emitted by the user at the corresponding location. In this case, the initial audio channel can be directly determined as the target audio channel, that is, the audio channel that needs to be woken up.

[0093] 507. If the first voiceprint comparison result indicates that the comparison fails, the audio channel corresponding to the reference voiceprint feature that passes the comparison with the real-time voiceprint feature, as indicated in the second voiceprint comparison result, shall be determined as the target audio channel.

[0094] In this embodiment of the application, if the first voiceprint comparison result indicates that the comparison fails, it means that the audio data of the initial audio channel is not the voice of the user at the corresponding position. Then, it is necessary to find the reference voiceprint features corresponding to other audio channels in the voiceprint feature database for comparison. If the reference voiceprint features of another audio channel are detected to pass the comparison with the real-time voiceprint features, then it can be considered that the user at the corresponding position of another audio channel spoke at the corresponding position of the initial audio channel. In order to more accurately wake up the audio channel of the speaking user, the audio channel corresponding to the reference voiceprint features that pass the comparison with the real-time voiceprint features can be determined as the target audio channel.

[0095] For example, suppose that both the driver's seat and the passenger seat are occupied by users, and a voiceprint feature database has been created based on the audio data of the driver and passenger users. The database stores the correspondence between the driver user, the driver's seat position, the driver's seat audio channel, and the driver user's baseline voiceprint features, as well as the correspondence between the passenger user, the passenger's seat position, the passenger's seat audio channel, and the passenger user's baseline voiceprint features. At this point, the passenger's body is leaning towards the driver's side, and their speech includes a preset wake-up message. The microphone array on the driver's side collects the passenger's audio data, and after model detection, it is determined that a wake-up command exists. The driver's side audio channel becomes the initial audio channel. Then, the real-time voiceprint features of the passenger's audio data are compared with the baseline voiceprint features of the driver's side audio channel. Obviously, the comparison fails. Next, the real-time voiceprint features of the passenger's audio data are compared with the baseline voiceprint features of other audio data stored in the voiceprint feature database. This time, the comparison with the baseline voiceprint features of the passenger's side audio channel stored in the voiceprint feature database passes. Therefore, the passenger's side audio channel is determined as the target audio channel, and the final wake-up is also the passenger's side audio channel.

[0096] 508. Wake up the target audio channel.

[0097] In this embodiment, relying on the multi-channel dedicated voiceprint registration and comparison error correction mechanism, the problem of multi-channel false wake-up in complex vehicle noise and reverberation environments is effectively solved, which greatly improves the accuracy and anti-interference ability of multi-zone voice wake-up in the whole vehicle, and improves the accuracy and stability of multi-zone wake-up.

[0098] Figure 6 This is a schematic diagram of the structure of a voice wake-up device provided in an embodiment of the present invention. The voice wake-up device can be an electronic device as described in the above embodiment, or it can be a component or assembly within that electronic device. Figure 6As shown, the voice wake-up device may include an acquisition module 601 and a processing module 602.

[0099] The acquisition module 601 is used to continuously acquire audio data corresponding to multiple audio channels in the vehicle through a multi-channel microphone array, with each audio channel corresponding to one microphone array. Processing module 602 is used to establish a voiceprint feature database based on the audio data corresponding to multiple audio channels. The voiceprint feature database includes the reference voiceprint features corresponding to each audio channel. The processing module 602 is also used to compare the real-time voiceprint features of the audio data of the initial audio channel with the voiceprint feature database when the target wake-up detection model detects that there is a wake-up command in the audio data of the initial audio channel, and obtain the voiceprint comparison result. The processing module 602 is also used to determine the target audio channel based on the voiceprint comparison results and to wake up the target audio channel.

[0100] Optionally, the acquisition module 601 is specifically used to acquire raw mixed voice data in the vehicle through a multi-microphone array; The processing module 602 is specifically used to separate the original speech mixture data into audio data by means of an audio decoupling algorithm, so as to obtain audio data corresponding to multiple audio channels respectively.

[0101] Optionally, the processing module 602 is specifically used to determine that the audio channel is a valid channel when a human voice signal is detected in the audio data corresponding to the audio channel; Processing module 602 is specifically used to extract voiceprint features from the voice signal of the effective channel to obtain the reference voiceprint features corresponding to the effective channel; The processing module 602 is specifically used to summarize the baseline voiceprint features of all valid channels and establish a voiceprint feature database.

[0102] Optionally, the processing module 602 is also used to filter the audio data of multiple audio channels, determine an initial audio channel from the multiple audio channels, and the audio data of the initial audio channel contains valid human voices; The processing module 602 is also used to perform content recognition on the audio data of the initial audio channel to obtain the audio content corresponding to the initial audio channel; The processing module 602 is also used to detect the audio content corresponding to the initial audio channel through the target wake-up detection model, and when the preset wake-up content is detected in the audio content corresponding to the initial audio channel, it is determined that there is a wake-up command in the audio data of the initial audio channel.

[0103] Optionally, the processing module 602 is specifically used to detect the audio content corresponding to the initial audio channel through the first wake-up detection model to obtain the first detection result; The processing module 602 is specifically used to detect the audio content corresponding to the initial audio channel again through the second wake-up detection model when the first detection result indicates that there is preset wake-up content in the audio content corresponding to the initial audio channel, so as to obtain the second detection result. The processing module 602 is specifically used to determine that there is a wake-up command in the audio data of the initial audio channel when the second detection result indicates that there is still preset wake-up content in the audio content corresponding to the initial audio channel, and the model parameters of the first wake-up detection model are less than the model parameters of the second wake-up detection model.

[0104] Optionally, the processing module 602 is specifically used to extract the reference voiceprint features corresponding to the initial audio channel from the voiceprint feature database when a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model. The processing module 602 is specifically used to compare the real-time voiceprint features of the audio data of the initial audio channel with the reference voiceprint features corresponding to the initial audio channel to obtain the first voiceprint comparison result. The processing module 602 is specifically used to compare the real-time voiceprint features of the audio data of the initial audio channel with the reference voiceprint features of other audio channels in the voiceprint feature database when the first voiceprint comparison result indicates that the comparison fails, so as to obtain the second voiceprint comparison result.

[0105] Optionally, the processing module 602 is specifically used to determine the initial audio channel as the target audio channel when the first voiceprint comparison result indicates that the comparison has passed; The processing module 602 is specifically used to determine the audio channel corresponding to the reference voiceprint feature that passed the comparison with the real-time voiceprint feature, as indicated in the second voiceprint comparison result, as the target audio channel when the first voiceprint comparison result indicates that the comparison fails.

[0106] Optionally, the processing module 602 is specifically used to compare the real-time voiceprint features of the audio data of the initial audio channel with the reference voiceprint features corresponding to the initial audio channel, and determine the first voiceprint similarity between the real-time voiceprint features and the reference voiceprint features corresponding to the initial audio channel. The processing module 602 is specifically used to determine that the comparison result of the first voiceprint is passed when the similarity of the first voiceprint is detected to be greater than or equal to a preset similarity threshold. The processing module 602 is specifically used to determine that the comparison result of the first voiceprint is unsuccessful when the similarity of the first voiceprint is less than a preset similarity threshold.

[0107] Figure 6 The voice wake-up device of the illustrated embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and will not be repeated here.

[0108] Figure 7 This is a schematic diagram of an electronic device according to an embodiment of the present invention. See below for details. Figure 7 It shows a structural schematic diagram of the electronic device 700 used to implement the embodiments of the present invention. Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0109] like Figure 7 As shown, the electronic device 700 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 into a random access memory (RAM) 703 to implement the project management method as described in the embodiments of the present invention. The RAM 703 also stores various programs and data required for the operation of the electronic device 700. The processing device 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0110] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic device 700 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 An electronic device 700 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0111] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts, thereby implementing the project management method described above. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 709, or installed from a storage device 708, or installed from a ROM 702. When the computer program is executed by the processing device 701, it performs the functions defined in the methods of the embodiments of the present invention.

[0112] It should be noted that the computer-readable medium described above in this invention can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0113] In addition, embodiments of the present invention also provide a vehicle, including: Figure 6 The vehicle is equipped with the voice wake-up device shown, which is used to implement the technical solution of the above method embodiment.

[0114] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0115] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0116] Optionally, when one or more of the above-described procedures are executed by the electronic device, the electronic device may also execute other steps of the above embodiments.

[0117] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0118] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0119] The units described in the embodiments of the present invention can be implemented in software or in hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0120] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0121] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0122] The above embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention.

Claims

1. A voice wake-up method, characterized in that, The method includes: The system continuously acquires audio data corresponding to multiple audio channels in the vehicle through a multi-channel microphone array, with each audio channel corresponding to one microphone array. A voiceprint feature database is established based on the audio data corresponding to the multiple audio channels, and the voiceprint feature database includes the baseline voiceprint features corresponding to each audio channel. When a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the real-time voiceprint features of the audio data of the initial audio channel are compared with the voiceprint feature database to obtain the voiceprint comparison result. Based on the voiceprint comparison results, the target audio channel is determined and the target audio channel is activated.

2. The method according to claim 1, characterized in that, The method of continuously acquiring audio data corresponding to multiple audio channels in the vehicle through a multi-channel microphone array includes: The original mixed voice data in the vehicle is collected through the multi-microphone array; The original speech mixture data is separated by an audio decoupling algorithm to obtain the audio data corresponding to the multiple audio channels.

3. The method according to claim 1, characterized in that, The step of establishing a voiceprint feature database based on the audio data corresponding to the multiple audio channels includes: When a human voice signal is detected in the audio data corresponding to the audio channel, the audio channel is determined to be a valid channel; Voiceprint features are extracted from the human voice signal of the effective channel to obtain the reference voiceprint features corresponding to the effective channel; The baseline voiceprint features of all valid channels are summarized to establish the voiceprint feature database.

4. The method according to claim 1, characterized in that, Before comparing the real-time voiceprint features of the initial audio channel with the voiceprint feature database to obtain the voiceprint comparison result when a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the method further includes: The audio data from the multiple audio channels are filtered to determine the initial audio channel, wherein the audio data of the initial audio channel contains valid human voices; Content recognition is performed on the audio data of the initial audio channel to obtain the audio content corresponding to the initial audio channel; The target wake-up detection model is used to detect the audio content corresponding to the initial audio channel. When the preset wake-up content is detected in the audio content corresponding to the initial audio channel, it is determined that the wake-up command exists in the audio data of the initial audio channel.

5. The method according to claim 4, characterized in that, The step of detecting the audio content corresponding to the initial audio channel using the target wake-up detection model, and determining that the wake-up command exists in the audio data of the initial audio channel when a preset wake-up content is detected in the audio content corresponding to the initial audio channel, includes: The first wake-up detection model is used to detect the audio content corresponding to the initial audio channel to obtain the first detection result. When the first detection result indicates that the preset wake-up content exists in the audio content corresponding to the initial audio channel, the audio content corresponding to the initial audio channel is detected again by the second wake-up detection model to obtain the second detection result; When the second detection result indicates that the preset wake-up content still exists in the audio content corresponding to the initial audio channel, it is determined that the wake-up instruction exists in the audio data of the initial audio channel, and the model parameters of the first wake-up detection model are less than the model parameters of the second wake-up detection model.

6. The method according to claim 1, characterized in that, When a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the real-time voiceprint features of the audio data of the initial audio channel are compared with the voiceprint feature database to obtain a voiceprint comparison result, including: When the wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, the reference voiceprint features corresponding to the initial audio channel are extracted from the voiceprint feature database; The real-time voiceprint features of the audio data of the initial audio channel are compared with the reference voiceprint features corresponding to the initial audio channel to obtain the first voiceprint comparison result. If the first voiceprint comparison result indicates that the comparison fails, the real-time voiceprint features of the audio data of the initial audio channel are compared with the baseline voiceprint features of other audio channels in the voiceprint feature database to obtain the second voiceprint comparison result.

7. The method according to claim 6, characterized in that, The step of determining the target audio channel based on the voiceprint comparison result includes: If the first voiceprint comparison result indicates that the comparison is successful, the initial audio channel is determined as the target audio channel; If the first voiceprint comparison result indicates that the comparison fails, the audio channel corresponding to the reference voiceprint feature that passed the comparison with the real-time voiceprint feature, as indicated in the second voiceprint comparison result, is determined as the target audio channel.

8. The method according to claim 6, characterized in that, The step of comparing the real-time voiceprint features of the audio data of the initial audio channel with the reference voiceprint features corresponding to the initial audio channel to obtain the first voiceprint comparison result includes: The real-time voiceprint features of the audio data of the initial audio channel are compared with the reference voiceprint features corresponding to the initial audio channel to determine the first voiceprint similarity between the real-time voiceprint features and the reference voiceprint features corresponding to the initial audio channel. When the similarity of the first voiceprint is detected to be greater than or equal to a preset similarity threshold, the first voiceprint comparison result is determined to be a successful comparison. When the similarity of the first voiceprint is detected to be less than the preset similarity threshold, the comparison result of the first voiceprint is determined to be a failure.

9. A voice wake-up device, characterized in that, The voice wake-up device includes: The acquisition module is used to continuously acquire audio data corresponding to multiple audio channels in the vehicle through a multi-microphone array, with each audio channel corresponding to one microphone array. The processing module is used to establish a voiceprint feature database based on the audio data corresponding to the multiple audio channels respectively. The voiceprint feature database includes the reference voiceprint features corresponding to each audio channel. The processing module is further configured to, when a wake-up command is detected in the audio data of the initial audio channel by the target wake-up detection model, compare the real-time voiceprint features of the audio data of the initial audio channel with the voiceprint feature database to obtain a voiceprint comparison result. The processing module is further configured to determine the target audio channel based on the voiceprint comparison result and to wake up the target audio channel.

10. An electronic device, characterized in that, include: Memory; processor; as well as Computer programs; The computer program is stored in the memory and configured to be executed by the processor to implement the voice wake-up method as described in any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the voice wake-up method as described in any one of claims 1-8.

12. A vehicle, characterized in that, Includes the voice wake-up device as described in claim 9.

Images