Method and system for configuring customized wake-up word of vehicle-mounted terminal screen based on voice control

Abstract

The invention relates to the technical field of voice recognition, and discloses a voice-control-based customized wake-up word configuration method and system for an in-vehicle machine screen, and the method comprises the steps: collecting the in-vehicle noise data of a target vehicle, and constructing a noise database; dividing noise records in the noise database into a plurality of noise levels; obtaining a user-defined wake-up word voice sample, and extracting a voiceprint feature vector of the user-defined wake-up word; measuring the similarity between the voiceprint feature vector of the custom wake-up word and each voiceprint feature vector in a voiceprint feature library, and matching to obtain a first voiceprint feature with the highest similarity; constructing a wake-up word detection sub-model corresponding to the noise level, collecting audio data in the vehicle in real time, performing wake-up word recognition according to the wake-up word detection sub-model and the first voiceprint feature, and triggering a vehicle machine screen to wake up; according to the invention, the accuracy and response speed of wake-up word recognition are improved.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and more specifically, to a method and system for configuring a custom wake-up word for a vehicle screen based on voice control. Background Art

[0002] With the rapid development of artificial intelligence technology, voice interaction has become one of the important ways of human-computer interaction. Especially in the field of intelligent driving, controlling vehicle-mounted devices through voice commands can maximize driving safety. In-vehicle voice assistants can complete tasks such as navigation, music playback, and air conditioning control through voice interaction while the driver focuses on the road conditions. Among them, voice wake-up is a key link in realizing in-vehicle voice interaction.

[0003] In the existing in-vehicle voice wake-up system, the wake-up is mainly triggered by a preset wake-up word or a user-defined wake-up word. For example, the patent application with publication number CN113611294A discloses a voice wake-up method, which configures multiple combined wake-up words and performs wake-up when the user voice matches the combined wake-up word. This method supports preset wake-up words, custom wake-up words, and multiple wake-up words, but does not consider the impact of complex acoustic environments such as in-car noise on the wake-up performance. For example, the patent with authorization announcement number CN109360552B proposes a method for automatically filtering wake-up words. By comparing the user voice with the wake-up word audio, meaningless wake-up words in the user voice are obtained and shielded, thereby improving the accuracy of semantic parsing. This method focuses on the semantic understanding stage after wake-up, but is prone to false wake-ups in complex acoustic environments.

[0004] In summary, the existing voice wake-up technology has not fully considered the impact of in-car noise on wake-up performance, and the custom wake-up words do not match the user's voiceprint sufficiently; the in-car noise is time-varying and diverse, and the wake-up system needs to maintain stable wake-up performance in various complex noise environments, while the existing methods generally lack an adaptation mechanism for in-car noise; factors such as user timbre, speaking speed, and emotions will affect the acoustic characteristics of custom wake-up words. If there is a large difference between the custom wake-up word model and the user's actual pronunciation, the wake-up accuracy will decrease; the existing custom wake-up word methods usually only train a unified acoustic model and lack the ability to model the user's personalized voiceprint features. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for configuring a custom wake-up word for a car screen based on voice control. The method first collects the noise data in the car to build a noise database, and divides the noise records into multiple noise levels; then obtains the custom wake-up word voice sample, matches the user's voiceprint features; and then builds a wake-up word detection sub-model for different noise levels. In the wake-up recognition stage, according to the noise level to which the in-car audio data belongs, the corresponding wake-up word detection sub-model is selected, and the user's voiceprint features are combined for matching verification, so as to accurately identify the custom wake-up word in a complex in-car noise environment, significantly improving the wake-up accuracy and system response speed.

[0006] To achieve the above object, the present invention provides the following technical solutions: The method for configuring a custom wake-up word for a car screen based on voice control includes: Collect the in-vehicle noise data of the target vehicle and build an in-vehicle noise database, wherein the in-vehicle noise database includes n1 noise records; divide the n1 noise records in the in-vehicle noise database into N1 noise levels, and record the noise level label of each noise record; obtain the user's custom wake-up word voice sample, extract the voiceprint feature vector of the custom wake-up word voice sample, and mark it as the custom wake-up word voiceprint feature vector; measure the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in the pre-built voiceprint feature library, and match the first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature vector; According to the in-car noise database and the first voiceprint feature, a wake-up word detection sub-model corresponding to N1 noise levels is constructed; Collect in-car audio data in real time, recognize the wake-up word based on the in-car audio data, the wake-up word detection sub-model and the first voiceprint feature, and trigger the car screen to wake up.

[0007] Furthermore, each of the n1 noise records includes attribute data of a noise segment; the attribute data of the noise segment includes a noise segment number, noise segment data, a noise type label, a noise energy value and a noise spectrum feature vector.

[0008] The dividing the n1 noise records in the in-vehicle noise database into N1 noise levels includes: clustering the n1 noise records according to the noise energy values ​​and noise spectrum feature vectors of the n1 noise records in the in-vehicle noise database.

[0009] Furthermore, clustering the n1 noise records includes: Step S1210, calculating the energy mean E according to the noise energy value and noise spectrum feature vector of each noise record in the vehicle noise database i , spectrum center F i , and the spectral dispersion Di , which constitutes the noise acoustic feature vector [E i ,F i ,D i ]; among them, E i is the energy mean of the ith noise record, F i is the spectral center of the ith noise record, D i is the spectral dispersion of the ith noise record, [E i ,F i ,D i ] represents the noise acoustic feature vector of the i-th noise record; Step S1220, using the noise acoustic feature vector [E i ,F i ,D i ] is the feature description, clustering n1 noise records to obtain N1 noise clustering centers; Step S1230, calculate the average silhouette coefficient SC, if SC≤SC', adjust the number of noise cluster centers N1, return to step S1220 for re-clustering until SC>SC', and output the clustering result; SC' is a preset silhouette coefficient threshold.

[0010] Furthermore, the method for constructing the voiceprint feature library is as follows: collecting voice samples of multiple users, extracting voiceprint feature vectors of the voice samples, and constructing a voiceprint feature library; the voiceprint feature vector includes fundamental frequency, formant and speech rate; The first voiceprint feature having the highest similarity to the custom wake-up word voiceprint feature vector comprises: The similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in the voiceprint feature library is measured to obtain a similarity score; the similarity scores are sorted, and the voiceprint feature vector in the voiceprint feature library with the highest score is selected and marked as the first voiceprint feature.

[0011] Furthermore, constructing the wake-up word detection sub-model corresponding to N1 noise levels includes: Based on the N1 noise levels in the in-vehicle noise database, N1 noise data subsets are constructed, each noise data subset corresponds to a noise level; Traverse N1 noise levels, and build an independent wake-up word detection sub-model for each noise level; obtain the wake-up word detection sub-model corresponding to all N1 noise levels, and the output of the wake-up word detection sub-model is a custom wake-up word detection result represented by a Boolean value, if it is 1, it means that the custom wake-up word is detected, if it is 0, it means that the custom wake-up word is not detected; The constructing of N1 noise data subsets includes: Traverse the n1 noise records in the in-vehicle noise database, and divide each noise record into corresponding N1 noise data subsets according to the noise level label.

[0012] Furthermore, for each noise level, constructing an independent wake-up word detection sub-model includes: According to the preset data set division ratio, the jth noise data subset is divided into a training set, a validation set, and a test set; 1≤j≤N1; The training set in the j-th noise data subset is merged with the first voiceprint feature to construct a wake-up word detection training set for noise level j; Taking the wake-up word detection training set for noise level j as input, train the initial wake-up word detection sub-model under noise level j.

[0013] Further, the performing of wake-up word recognition based on the in-vehicle audio data, the wake-up word detection sub-model and the first voiceprint feature to trigger the vehicle computer screen to wake up includes: Step S3100, determining the noise level to which the in-car audio data belongs, and selecting a wake-up word detection sub-model corresponding to the noise level to which the in-car audio data belongs to perform custom wake-up word detection, and determining whether the custom wake-up word is detected; Step S3200: If no custom wake-up word is detected, jump to step S3100 to continue the next round of in-vehicle audio data collection and detection; if a custom wake-up word is detected, the wake-up word audio data is collected through the microphone array, the wake-up word speaker is located, and the horizontal azimuth angle of the wake-up word speaker relative to the vehicle microphone array is obtained. and distance ; Step S3300: Setting the wake-up angle threshold range and distance threshold range ;like and , it is determined that the detected custom wake-up word comes from a reasonable position in the car and is identified as a wake-up word to be confirmed; otherwise, it is considered as a false wake-up, and the process jumps to step S3100 to continue the next round of in-car audio data collection and custom wake-up word detection; Step S3400, extract acoustic features from the wake-up word audio data determined to be the wake-up word to be confirmed, and construct a second voiceprint feature; calculate the similarity score SIM between the second voiceprint feature and the first voiceprint feature. If the similarity score SIM is greater than the preset voiceprint verification threshold, the wake-up word to be confirmed is identified as a valid wake-up word, triggering the car screen to wake up; otherwise, it is identified as an invalid wake-up word, rejects the car screen wake-up and jumps to step S3100 to continue the next round of in-car audio data collection and custom wake-up word detection.

[0014] Further, the determining of the noise level of the in-vehicle audio data includes: Extract features from the real-time collected in-car audio data to obtain the real-time energy value Es, the real-time spectrum center of gravity Fs and the real-time spectrum dispersion Ds, and construct the real-time acoustic feature vector [Es, Fs, Ds]; The real-time acoustic feature vector [Es, Fs, Ds] is compared with N1 noise cluster centers, the Euclidean distance between the real-time acoustic feature vector and each noise cluster center is calculated, and the noise level corresponding to the noise cluster center with the smallest Euclidean distance is selected as the noise level to which the real-time collected in-vehicle audio data belongs.

[0015] Furthermore, the horizontal azimuth angle of the wake-up word speaker relative to the vehicle microphone array is obtained. and distance include: The wake-up word audio data is collected by a microphone array. The wake-up word audio data has M voice signals in total. The wake-up word audio data of the mth voice signal is recorded as , t is time, ; Wake-up word audio data Perform voice endpoint detection and extract the voice received by each microphone. Find the start and end time of the wake-up word voice segment in , the wake-up word speech segment is recorded as ; Select one microphone from the microphone array as the reference microphone, and the other M-1 microphones as non-reference microphones, and estimate the time delay of the wake-up word speech segment between the non-reference microphone and the reference microphone ,in , It is The time delay of the wake-up word speech segment of the non-reference microphone relative to the reference microphone; based on the geometric layout of the microphone array, construct M-1 equations:

[0016] in It is The position vectors of the non-reference microphones relative to the reference microphone, is the speed of sound, is the horizontal azimuth of the sound source, is the pitch angle of the sound source; Solve the M-1 equations simultaneously to obtain the horizontal azimuth angle of the wake-up word speaker relative to the microphone array: and distance ,in .

[0017] A vehicle screen custom wake-up word configuration system based on voice control, which is used to implement the above-mentioned vehicle screen custom wake-up word configuration method based on voice control, the system includes: Noise level classification module: used to collect the in-vehicle noise data of the target vehicle and build an in-vehicle noise database, wherein the in-vehicle noise database includes n1 noise records; and classify the n1 noise records in the in-vehicle noise database into N1 noise levels; The first voiceprint feature acquisition module is used to obtain the user's custom wake-up word voice sample, extract the voiceprint feature vector of the custom wake-up word voice sample, and mark it as the custom wake-up word voiceprint feature vector; measure the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in the pre-built voiceprint feature library, and match to obtain the first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature vector; Model building module: used to build a wake-up word detection sub-model corresponding to N1 noise levels based on the in-vehicle noise database and the first voiceprint feature; Wake-up word recognition module: used to collect in-car audio data in real time, perform wake-up word recognition based on in-car audio data, wake-up word detection sub-model and first voiceprint features, and trigger the car screen to wake up.

[0018] An electronic device includes a memory, a central processing unit, and a computer program stored in the memory and executable on the central processing unit. When the central processing unit executes the computer program, the method for configuring a custom wake-up word for a vehicle screen based on voice control is implemented.

[0019] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the above-mentioned method for configuring a custom wake-up word for a vehicle screen based on voice control.

[0020] Compared with the prior art, the present invention has the following beneficial effects: Improve recognition accuracy. By building a noise level database and a custom wake-up word voiceprint feature library, the wake-up word can be accurately identified in different noise environments, and the acoustic feature vector and similarity matching algorithm are used to effectively reduce the false wake-up rate. Enhance system response speed, collect and process in-car audio data in real time, quickly determine the wake-up word, and achieve instant response. At the same time, use an independent wake-up word detection sub-model to optimize for different noise levels to improve detection efficiency. Adapt to diverse environments. Through clustering analysis of noise data, the system can adapt to noise changes in various driving environments, and design a flexible noise level division mechanism so that the system can maintain stable performance in complex acoustic environments. Improve user experience, allow users to customize wake-up words, improve personalization and convenience, and avoid false wake-ups and invalid operations through accurate sound source positioning, and enhance interactive experience. Optimize resource usage. Through clustering and model optimization, reduce computing resource consumption, improve overall system efficiency, and achieve efficient noise data management and retrieval, which is convenient for subsequent analysis and optimization. These advantages make the present invention have significant technical breakthroughs and application value in the field of in-vehicle voice interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 The following is a flow chart showing the principle of the method for configuring a custom wake-up word for a vehicle screen based on voice control in the present invention; Figure 2 A flow chart of a method for clustering n1 noise records in a method for configuring a custom wake-up word for a vehicle screen based on voice control of the present invention; Figure 3 A flowchart of a method for constructing a wake-up word detection sub-model corresponding to N1 noise levels in the method for configuring a custom wake-up word for a vehicle screen based on voice control of the present invention; Figure 4 A flowchart of a method for constructing an independent wake-up word detection sub-model for each noise level in the method for configuring a custom wake-up word for a vehicle screen based on voice control of the present invention; Figure 5 A flowchart of a method for setting a wake-up angle threshold range and a distance threshold range in a method for configuring a custom wake-up word for a vehicle screen based on voice control of the present invention; Figure 6 This is a functional module diagram of the vehicle screen custom wake-up word configuration system based on voice control in the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Example 1 See also Figure 1 As shown, this embodiment provides a method for configuring a custom wake-up word for a vehicle screen based on voice control, including: Step S1000, collecting the in-vehicle noise data of the target vehicle, building an in-vehicle noise database, wherein the in-vehicle noise database includes n1 noise records; dividing the n1 noise records in the in-vehicle noise database into N1 noise levels; obtaining a user's custom wake-up word voice sample, measuring the similarity between the voiceprint feature of the custom wake-up word voice sample and each voiceprint feature in a preset voiceprint feature library, and matching to obtain a first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature; Further, step S1000 includes: Step S1100, collecting the in-vehicle noise data of the target vehicle and constructing an in-vehicle noise database; the in-vehicle noise database includes n1 noise records, each noise record includes attribute data of a noise segment; the attribute data of the noise segment includes a noise segment number, noise segment data, noise type label, noise energy value and noise spectrum feature vector; Specifically, step S1100 obtains the in-vehicle noise in the real environment through the vehicle-mounted acoustic sensor, which comprehensively covers all kinds of noise interference in the driving scene. First, the in-vehicle acoustic sensor array is used to collect the in-vehicle noise data of the target vehicle, including but not limited to engine noise, wind noise, external traffic noise, and human voice noise in the car; wherein the acoustic sensor array includes microphone arrays, MEMS sensors, etc., which are arranged in the driving area, passenger area and other locations in the car; wherein, engine noise reflects the working condition of the vehicle and is one of the main noise sources affecting voice interaction; wind noise is airflow noise caused by vehicle speed, which will increase with the increase of vehicle speed; external traffic noise includes noise from other vehicles, horn noise, etc., which is the background noise introduced by the external environment; human voice noise in the car comes from the conversation and breathing of passengers, etc., which is non-voice noise generated by activities in the car.

[0025] The collected in-car noise data is preprocessed, including noise segmentation, energy normalization, noise reduction filtering and other operations; noise segmentation is to divide the noise data into different noise segments using voice endpoint detection algorithms, such as double threshold method and energy entropy ratio method; energy normalization is to normalize the energy of each noise segment to make its amplitude range consistent; noise reduction filtering is to use adaptive filters, such as LMS and RLS, to reduce noise and enhance noise; through noise segmentation, noise can be extracted from the continuous audio stream, which is convenient for subsequent directional processing. Energy normalization eliminates the amplitude difference caused by factors such as noise recording equipment and distance, so that the noise data has a unified energy benchmark. While retaining the essential characteristics of noise, noise reduction filtering removes the random noise components and improves the signal-to-noise ratio. Based on the preprocessed in-car noise data, a structured noise library is constructed, which provides a unified data source for noise classification on the one hand, and makes noise data easier to retrieve, manage and apply on the other hand, providing a data basis for the development, testing and iteration of the wake-up system.

[0026] Based on the preprocessed in-car noise data, a structured in-car noise database is constructed; the noise segment number in the in-car noise database uniquely identifies each noise segment, which is convenient for data retrieval and management; such as 001, 002, which is used to distinguish different noise segments. The noise segment data is used to store the audio data of each noise segment for analysis and processing, such as an audio file containing 5 seconds of engine noise. The noise type label identifies the source or category of the noise, which is convenient for classification and processing, such as "engine noise", "wind noise", "traffic noise" and "human voice noise". The noise energy value represents the loudness or intensity of the noise, which is used for comparison and normalization processing, such as 85dB, which represents the sound pressure level of the noise. The noise spectrum feature vector describes the frequency characteristics of the noise, which is used for signal processing and pattern recognition. A spectrum feature vector shows the peak of a specific frequency to help identify the noise type. Therefore, step S1100 collects and optimizes the in-car noise data in a real environment, lays a solid data foundation for subsequent noise classification and model adaptation, and is the key to improving the environmental adaptability and robustness of the wake-up system.

[0027] Step S1200, clustering the n1 noise records in the in-vehicle noise database according to the noise energy values ​​and noise spectrum feature vectors of the n1 noise records, dividing the n1 noise records in the in-vehicle noise database into N1 noise levels, and recording the noise level label of each noise record; Further, such as Figure 2 As shown, step S1200 includes: Step S1210, calculating the energy mean E according to the noise energy value and noise spectrum feature vector of each noise record in the vehicle noise database i , spectrum center F i , and the spectral dispersion Di , which constitutes the noise acoustic feature vector [E i ,F i ,D i ]; among them, E i is the energy mean of the ith noise record, F i is the spectral center of the ith noise record, D i is the spectral dispersion of the ith noise record, [E i ,F i ,D i ] represents the noise acoustic feature vector of the i-th noise record; Step S1220, using the noise acoustic feature vector [E i ,F i ,D i ] is the feature description, clustering n1 noise records to obtain N1 noise clustering centers; Step S1230, calculate the average silhouette coefficient SC, if SC≤SC', adjust the number of noise cluster centers N1, return to step S1220 for re-clustering until SC>SC', and output the clustering result; SC' is a preset silhouette coefficient threshold.

[0028] The calculation of the average silhouette coefficient SC comprises:

[0029] in: : No. Noise cluster centers, The value range is 1 to N1.

[0030] : No. The cluster center number to which the noise record belongs.

[0031] : No. Noise Record and The Euclidean distance between cluster centers is used to measure the noise records. With cluster center The distance is calculated as: ; in Cluster centers The energy mean, spectral center of gravity and spectral dispersion of

[0032] molecular Indicates The closest distance between a noise record and other categories minus the distance between it and the center of the category to which it belongs. The larger the numerator, the farther the noise record is from the boundary of other categories, the higher the cohesion with the category to which it belongs, and the better the clustering effect.

[0033] Denominator This is to normalize the silhouette coefficients of different records. The larger value of the distance between a noise record and the center of the cluster it belongs to and its closest distance to other clusters. This can balance the impact of the internal distance of the cluster and the distance between clusters on the silhouette coefficient.

[0034] For all The contour coefficients of the noise records are averaged to obtain the average contour coefficient SC, which has a value range of When SC is closer to 1, the cohesion and separation of the clustering are better, and the rationality of the noise level division is higher. When SC is greater than the given threshold SC', it can be considered that the current noise clustering result is excellent.

[0035] The function of this formula is to quantitatively evaluate the effect of noise clustering. The closer the noise level number N1 is to the actual noise distribution, the closer the noise cluster center C is. k The more accurately the acoustic characteristics of each noise level are reflected, the higher the average contour coefficient SC will be. By iteratively adjusting the value of N1 to make SC reach a larger value, the noise clustering result can be made more reasonable, thereby providing better data support for subsequent applications such as hierarchical noise reduction and adaptive wake-up.

[0036] It is worth mentioning that since the silhouette coefficient calculates the relative distance between each noise record and its category and other categories, it is robust to the encoding of the order and category labels of the noise records. Even if the order of the noise records changes or the clustering categories are renumbered, the SC value remains unchanged as long as the clustering result remains unchanged. This helps to improve the reliability of clustering evaluation.

[0037] In summary, the average silhouette coefficient formula comprehensively considers the compactness and dispersion of noise clustering, and gives a quantitative evaluation of the rationality of noise classification. The average silhouette coefficient changes with the change of the number of cluster centers N1. When N1 gradually approaches the actual noise level, SC will reach a larger value. Furthermore, the value of N1 can be automatically adjusted through the parameter optimization algorithm to maximize SC, thereby achieving adaptive optimization of noise level classification. This formula is of great significance for the mining of in-vehicle noise data and the improvement of the environmental adaptability of in-vehicle voice interaction technology.

[0038] Specifically, the energy mean (E i) reflects the average energy level of the ith noise record. It is calculated by summing the squares of the noise energy values ​​at all times of the noise record and then dividing it by the length of the record. The larger the energy mean, the higher the loudness of the noise. The center of gravity of the spectrum (F i ) represents the center position of the spectrum distribution of the ith noise record. It is calculated by summing the product of the frequency of each frequency point and the square of its corresponding spectrum amplitude, and then dividing it by the sum of the square of the spectrum amplitude. The larger the spectrum center of gravity, the higher the frequency component of the noise. Spectral dispersion (D i ) is used to measure the dispersion of the spectrum distribution of the i-th noise record. It is obtained by calculating the sum of the products of the square of the difference between the frequency of each frequency point and the center of gravity of the spectrum and the square of the spectrum amplitude, and then dividing it by the sum of the square of the spectrum amplitude. The larger the spectrum dispersion, the more dispersed the noise spectrum is and the richer the frequency components it contains. For example, the energy mean of an engine noise record is 80 decibels, the center of gravity of the spectrum is 500 Hz, and the spectrum dispersion is 200 Hz squared, so its acoustic eigenvector can be expressed as [80,500,200]. The energy mean of a wind noise record is 60 decibels, the center of gravity of the spectrum is 2000 Hz, the spectrum dispersion is 1000 Hz squared, and the acoustic eigenvector is [60,2000,1000]. It can be seen that the engine noise has higher energy, lower frequency and is more concentrated, while the wind noise has lower energy but higher frequency and is dispersed.

[0039] These three features are chosen as the acoustic eigenvectors of noise because they fully describe the acoustic properties of noise in terms of energy, frequency, and dispersion. This feature description enables noise clustering to group noise records with similar energy levels and similar spectral distributions into one category, and the resulting noise levels have acoustic homogeneity.

[0040] Using algorithms such as K-means clustering, with noise acoustic feature vectors as sample features, multiple noise records are clustered and analyzed to obtain multiple cluster centers. Each cluster center represents a noise level, including the average energy, spectral center of gravity, and spectral dispersion of the noise at that level.

[0041] For example, the clustering results may yield three noise levels, with cluster centers being: The first level center is [70,800,300], which represents low-frequency noise, which may include engine noise, etc. The second level center is [65,1500,500], which represents medium-frequency noise, which may include traffic noise outside the car and human voice noise; The third level center is [50,4000,1500], which represents high-frequency noise, which may include wind noise, etc.

[0042] The clustering effect is evaluated by calculating the average silhouette coefficient. The silhouette coefficient is an indicator of the compactness and separation of clusters. Its value range is -1 to 1. The larger the value, the better the clustering quality. For each noise record, the silhouette coefficient is calculated by comparing the average distance of the record with the same type of noise and the average distance with the nearest heterogeneous noise. If the silhouette coefficient is close to 1, it means that the record is correctly classified; if it is close to -1, it may be misclassified; if it is close to 0, it means that the record is between two categories.

[0043] The average silhouette coefficient is the arithmetic mean of all silhouette coefficients, reflecting the overall quality of the clustering results. When the average silhouette coefficient is greater than the preset silhouette coefficient threshold, it can be considered that the clustering effect is ideal and the noise level division is reasonable. Otherwise, it is necessary to adjust the number of clusters and re-cluster until the ideal effect is achieved.

[0044] For example, if the average silhouette coefficient is 0.8, it means that the three current noise levels have high distinction in acoustic features and the clustering result is reliable. If the value is 0.3, you may need to try adjusting the number of noise levels (for example, to 2 or 4) and perform clustering again to find the best classification scheme.

[0045] Through such iterative optimization, a noise classification result with separable acoustic features was finally achieved. This result reveals the inherent structure and level division of in-vehicle noise data, providing an important basis for subsequent adaptive noise suppression and graded wake-up. Compared with the simple energy threshold classification method, this method makes full use of the frequency domain characteristics of noise, achieves more refined and targeted noise classification, and lays the foundation for improving the environmental adaptability of the in-vehicle voice interaction system.

[0046] In summary, step S1200 successfully achieves automatic classification of in-vehicle noise data by extracting and clustering noise acoustic features. This method combines signal processing and machine learning technologies, and can mine valuable acoustic patterns and rules from complex in-vehicle noise, providing data support and algorithm guarantee for intelligent and personalized in-vehicle voice interaction.

[0047] Step S1300, obtain the user's custom wake-up word voice sample, extract the voiceprint feature vector of the custom wake-up word voice sample, and mark it as the custom wake-up word voiceprint feature vector; measure the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature in the pre-built voiceprint feature library, and match the first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature vector.

[0048] Further, step S1300 includes: Step S1310, collecting voice samples of multiple users, extracting voiceprint feature vectors of the voice samples, and constructing a voiceprint feature library; the voiceprint feature vector includes fundamental frequency, formant and speech rate; Step S1320, measuring the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in the voiceprint feature library to obtain a similarity score; sorting according to the similarity score, selecting the voiceprint feature vector in the voiceprint feature library with the highest score, and marking it as the first voiceprint feature.

[0049] Specifically, step S1300 obtains the user's custom wake-up word voice sample, extracts its voiceprint features, and matches it with the preset voiceprint feature library to achieve personalized voiceprint registration of the custom wake-up word. First, the system prompts the user to record a voice sample of the custom wake-up word, such as "XX classmate", "Hello, Xiao X", etc. The user can use the car microphone to record the wake-up word voice multiple times in a quiet environment to obtain the purest voice data possible.

[0050] Step S1310 obtains rich voiceprint data by collecting voice samples from multiple users. The voice samples cover users of different genders, ages, and dialects, and include multiple types such as sentences and words, ensuring the diversity and representativeness of the voiceprint feature library. For each voice sample, the voiceprint feature vector is extracted using voiceprint analysis technology. The fundamental frequency is a measure of the vibration frequency of the vocal cords, reflecting the speaker's age, gender and other physiological characteristics. It can be extracted by the average amplitude difference function (AMDF) based on the time domain or the cepstrum method based on the frequency domain. The resonance peak is the frequency domain peak generated by the resonance of the vocal tract, reflecting the characteristics of the speaker's vocal organs. It can be estimated using methods such as linear predictive coding (LPC). Generally, the center frequency and bandwidth of the first 3-5 resonance peaks are extracted as features. The speech rate is the number of pronunciation units (such as syllables, words, etc.) per unit time, reflecting the speaker's rhythmic habits. It is obtained by endpoint detection and pronunciation unit division statistics of speech.

[0051] For example, the voiceprint features of a user's speech sample can be expressed as [103.5Hz, 650Hz, 1400Hz, 2600Hz, 3.5 syllables / second], which correspond to the five feature dimensions of fundamental frequency, the first three formants, and speech rate. By extracting features from the voice samples of multiple users, a voiceprint feature library that encompasses group differences can be constructed. This is the basis for voiceprint recognition and matching.

[0052] By extracting the voiceprint features of the speech samples, each user's voiceprint is converted into a feature vector of fixed dimensions for easy comparison and matching. The feature set composed of the voiceprint feature vectors of each user is the user voiceprint feature library. This feature library covers the voiceprint templates of different users and is the basis for voiceprint recognition and verification. Pre-building a voiceprint feature library can speed up voiceprint matching without repeated feature extraction. On the other hand, it is conducive to collecting more comprehensive user voice samples and improving the richness and reliability of voiceprint features.

[0053] The voiceprint similarity measurement method is used to calculate the similarity between the wake-up word voiceprint feature and the voiceprint feature of each user in the voiceprint library. Common similarity measurements include Euclidean distance, cosine similarity, and probability distance. Euclidean distance measures the linear distance between two voiceprint feature vectors in the feature space. The smaller the distance, the closer the voiceprints are. Cosine similarity calculates the cosine value of the angle between two voiceprint feature vectors. The larger the value, the more consistent the direction, that is, the more similar the voiceprints are. Probability distance, such as KL divergence, measures the difference in the probability distribution of two voiceprint features. The smaller the difference, the more the voiceprints match. Regardless of the measurement method used, a similarity score can be obtained to quantify the similarity between the wake-up word voiceprint and each voiceprint in the library. The wake-up word voiceprint is calculated for similarity with each voiceprint in the library in turn, and a similarity score vector is obtained. Sort by similarity score from high to low. The voiceprint with the highest score is the voiceprint that best matches the current wake-up word voiceprint, and the corresponding user identity is also identified as the current wake-up user. Through similarity sorting, the voiceprint matching result is more reliable and the false recognition rate is low. The matching method based on the existing voiceprint library avoids the tedious steps of repeatedly entering the user's voice. The user only needs to enter the wake-up word once to easily complete the voiceprint registration and recognition.

[0054] For example, the voiceprint feature of the user-defined wake-up word "Hello, Xiao X" is [105Hz, 720Hz, 1300Hz, 2400Hz, 4 syllables / second]. The similarity calculation result with the three features in the voiceprint library is [0.8, 0.6, 0.3]. The first voiceprint feature corresponding to the highest score of 0.8 [110Hz, 700Hz, 1350Hz, 2500Hz, 3.8 syllables / second] is selected as the best match and can be used for reference and optimization of subsequent wake-up word acoustic modeling.

[0055] Extracting the voiceprint features of user-defined wake-up words implants the user's personality from the source of voice input, which helps to improve the pertinence of wake-up word recognition. By matching the similarity with the voiceprint feature library, the most suitable acoustic reference can be automatically selected for different users, which is adaptive and reduces the workload of manual tuning. It combines multiple acoustic features such as fundamental frequency and resonance peak to more comprehensively reflect the voiceprint properties than a single feature, improve the robustness of voiceprint representation and matching, and reduce the recognition error rate. The voiceprint feature library can be continuously expanded and optimized by incorporating more user voice samples, making the matching reference benchmark richer, more reliable, and more scalable.

[0056] In short, step S1300 selects the optimal acoustic reference for the user's customized wake-up word through voiceprint feature extraction and similarity matching, which not only retains the user's personality but also takes into account the reliability of recognition. It is an intelligent and personalized solution in the voice interaction system. It uses voiceprint recognition technology to implant a voiceprint adaptation mechanism in the wake-up word setting link, laying the foundation for subsequent wake-up word modeling and recognition, and is of great significance to improving system performance and user experience. Compared with the traditional "one-size-fits-all" wake-up word setting, this method is more flexible and intelligent, showing the development trend and potential of voice interaction technology.

[0057] Step S2000, constructing a wake-up word detection sub-model corresponding to N1 noise levels according to the in-vehicle noise database and the first voiceprint feature; Furthermore, if Figure 3 As shown, step S2000 includes: Step S2100, based on N1 noise levels in the in-vehicle noise database, construct N1 noise data subsets, each noise data subset corresponding to a noise level; The constructing of N1 noise data subsets includes: Traverse the n1 noise records in the in-vehicle noise database, and divide each noise record into corresponding N1 noise data subsets according to the noise level label of each noise record; Step S2200, constructing an independent wake-up word detection sub-model for each noise level; Furthermore, if Figure 4 As shown, step S2200 includes: Step S2210, dividing the jth noise data subset into a training set, a validation set, and a test set according to a preset data set division ratio; 1≤j≤N1; Step S2220, fusing the training set in the j-th noise data subset with the first voiceprint feature to construct a wake-up word detection training set for noise level j; Step S2230, taking the wake-up word detection training set for noise level j as input, training the initial wake-up word detection sub-model under noise level j; Step S2240, using the validation set in the j-th noise data subset, optimizing the model parameters to obtain the final wake-up word detection sub-model under the noise level j; Step S2250, using the test set in the j-th noise data subset, evaluates the performance indicators of the final wake-up word detection sub-model under the noise level j.

[0058] Step S2300, traverse N1 noise levels, execute step S2200, and obtain the wake-up word detection sub-model corresponding to all N1 noise levels. The output of the wake-up word detection sub-model is a custom wake-up word detection result represented by a Boolean value. If it is 1, it means that the custom wake-up word is detected. If it is 0, it means that the custom wake-up word is not detected.

[0059] Specifically, step S2000 aims to construct multiple specialized wake-up word detection sub-models for the wake-up word recognition problem in different noise environments, thereby improving the environmental adaptability and recognition performance of the system.

[0060] In step S2100, based on the N1 noise levels obtained previously, the original in-vehicle noise database is divided into N1 noise data subsets. Each subset corresponds to a specific noise type or noise intensity, such as "high-speed wind noise", "medium engine noise", etc. By grouping the n1 noise records according to the noise level label, N1 data subsets with consistent internal noise characteristics and high external distinction are finally obtained. This data set division method allows subsequent model training to be more targeted, which is conducive to improving the granularity and accuracy of wake-up word detection.

[0061] The core of step S2200 is to train a wake-up word detection sub-model for each noise level. Step S2210 first uses the holdout method to divide each noise data subset into a training set, a validation set, and a test set in a certain ratio (such as 7:2:1). Among them, the training set is used to train model parameters, the validation set is used to adjust parameters and select the optimal model, and the test set is used to evaluate the actual performance of the model. This data set division helps to improve the generalization ability and robustness of the model.

[0062] Step S2220 aims to fuse the custom wake-up word voiceprint features registered by the user with the noise data to construct a personalized wake-up word detection training set. Specifically, the user's custom wake-up word voiceprint feature vector is spliced ​​with each noise record in the training set to form a new training sample. For example, a training sample that originally contained 5 seconds of noise becomes a new 45-dimensional feature vector after being spliced ​​with the extracted 40-dimensional MFCC voiceprint features. This feature fusion enables the training data to contain both the acoustic information of the ambient noise and the target wake-up word, so that the trained model can accurately identify the user's wake-up word voice in complex noise. Compared with the traditional wake-up word detection model, this method makes full use of the personalized characteristics of the user's voiceprint and significantly improves the sensitivity and accuracy of wake-up.

[0063] Step S2230 uses a combined model of CNN and LSTM to train the wake-up word detection sub-model. CNN (convolutional neural network) can automatically extract local features in speech signals, and LSTM (long short-term memory network) can model long-term dependencies of speech. The combination of the two can comprehensively characterize the acoustic and temporal characteristics of the wake-up word speech. The input of the model is a noisy speech segment that incorporates voiceprint features, and the output is a custom wake-up word detection result represented by a Boolean value. During training, the manually annotated wake-up word segments are used as positive samples, and other speech segments and pure noise segments are used as negative samples. The cross-entropy loss function between the sample prediction results and the true labels is minimized, and the model parameters are iteratively optimized. After multiple rounds of training, the model can accurately locate and identify the target wake-up word from a noisy audio stream.

[0064] For example, a positive sample in the training set may be a 3-second voice of a user saying "Hello, Xiao X" in wind noise of about 80 decibels, which is concatenated with the user's voiceprint feature and input into the model training. The model outputs "1", which means that the custom wake-up word is detected, which is consistent with the manual annotation, indicating that the model has learned to recognize the user's wake-up word sound in moderate wind noise.

[0065] Step S2240 uses the validation set to tune the model. By traversing different hyperparameters (such as the convolution kernel size of CNN and the number of hidden layer units of LSTM) through grid search and other methods, the parameter combination with the best performance on the validation set is selected, and the early stopping method is used to prevent overfitting, and finally a wake-up word detection sub-model with stable performance and good generalization is obtained. This process ensures that the optimization direction of the model is consistent with the performance in the real environment, reducing the difference between the training set and the actual application.

[0066] Step S2250 uses the test set to evaluate the performance indicators of the model, such as accuracy, recall, false awakening rate, etc. By analyzing these indicators, the actual performance of the wake-up word detection sub-model under the noise level can be comprehensively judged, providing a basis for subsequent model selection and improvement. An ideal sub-model should have high target wake-up word recognition accuracy and low false awakening rate, while also being able to adapt to changes in the acoustic environment within the noise level.

[0067] After iterations of step S2300, N1 wake-up word detection sub-models for different noise levels are finally obtained. Each sub-model is trained with special noise data and can robustly detect the user's customized wake-up word in a specific in-car noise environment. Compared with a single general wake-up model, this divide-and-conquer, tailored multi-model mechanism has obvious advantages in complex noise response and personalized wake-up word recognition.

[0068] In summary, step S2000 constructs an adaptive wake-up word detection solution in a vehicle environment through key technologies such as noise data set division, voiceprint feature fusion, and hierarchical modeling. This solution can dynamically select the optimal wake-up word detection sub-model according to the real-time changes in the noise in the car, thereby achieving a good balance in noise suppression, wake-up sensitivity, false wake-up rate, etc., laying a solid foundation for the practical application of the in-vehicle voice interaction system.

[0069] Step S3000: collect in-car audio data in real time, perform wake-up word recognition based on the in-car audio data, the wake-up word detection sub-model and the first voiceprint feature, and trigger the car computer screen to wake up.

[0070] Furthermore, step S3000 includes: Step S3100, collecting in-car audio data in real time, determining the noise level to which the in-car audio data belongs, and selecting a wake-up word detection sub-model corresponding to the noise level to which the in-car audio data belongs to perform custom wake-up word detection, and determining whether the custom wake-up word is detected; Further, step S3100 includes: Step S3110, extracting features from the in-car audio data collected in real time, obtaining a real-time energy value Es, a real-time spectrum center of gravity Fs and a real-time spectrum dispersion Ds, and constructing a real-time acoustic feature vector [Es, Fs, Ds]; Step S3120, comparing the real-time acoustic feature vector [Es, Fs, Ds] with N1 noise cluster centers, calculating the Euclidean distance between the real-time acoustic feature vector and each noise cluster center, and selecting the noise level corresponding to the noise cluster center with the smallest Euclidean distance as the noise level to which the real-time collected in-vehicle audio data belongs; Step S3130: According to the noise level of the in-car audio data, a corresponding wake-up word detection sub-model is selected to perform customized wake-up word detection.

[0071] Specifically, step S3100 uses the vehicle-mounted microphone array to continuously collect in-car audio signals to achieve real-time monitoring of in-car audio. The vehicle-mounted microphone array consists of multiple microphone units distributed in different positions in the car to pick up all-round in-car sounds. The audio data collected in real time is processed in frames, each frame contains L sampling points, and a certain proportion of overlap is allowed between adjacent frames to smooth the time changes of audio features. For example, the audio signal is collected at a sampling rate of 16kHz, the frame length is L=400 sampling points (corresponding to 25ms), and the frame shift is 160 sampling points (corresponding to 10ms), then the overlap between adjacent frames is 50%.

[0072] The audio frames are pre-processed by pre-emphasis, framing, and windowing to extract the features of the audio frames and obtain the real-time energy value Es, the real-time spectrum center of gravity Fs, and the real-time spectrum dispersion Ds. Pre-emphasis is to filter the audio signal with a first-order high-pass filter to enhance the high-frequency components and compensate for the high-frequency attenuation of the channel and the microphone. Framing is to divide the pre-emphasized audio signal frame by frame according to the frame length L to obtain a series of audio frames. Windowing is to apply smoothing window functions such as Hamming window and Hanning window to each frame of audio to reduce signal mutations at the edge of the frame. After pre-processing, the noise components in the audio frame are suppressed and the speech components are enhanced.

[0073] Step S3100 realizes real-time noise estimation and adaptive wake-up of in-car audio through feature extraction and noise classification, which is the key to improving the environmental adaptability of the in-vehicle voice wake-up system. This step combines the noise modeling of step S1000 and the hierarchical wake-up of step S2000 to dynamically optimize the wake-up word detection process in a complex noise environment, thereby maximizing the accuracy and real-time performance of the wake-up system. Compared with the traditional fixed model wake-up method, the noise hierarchical adaptive wake-up mechanism introduced in this step can significantly improve the wake-up quality in harsh noise environments and achieve more natural and smooth human-computer voice interaction.

[0074] Step S3200: If no custom wake-up word is detected, jump to step S3100 to continue the next round of in-vehicle audio data collection and detection; if a custom wake-up word is detected, the wake-up word audio data is collected through the microphone array, the wake-up word speaker is located, and the horizontal azimuth angle of the wake-up word speaker relative to the vehicle microphone array is obtained. and distance ; Further, step S3200 includes: Step S3210: If a custom wake-up word is detected, the wake-up word audio data is collected through the microphone array. The wake-up word audio data has a total of M voice signals, and the wake-up word audio data of the mth voice signal is recorded as , t is time, ; Step S3220: wake-up word audio data Perform voice endpoint detection and extract the voice received by each microphone. Find the start and end time of the wake-up word voice segment in , the wake-up word speech segment is recorded as ; Select one microphone from the microphone array as the reference microphone, and the other M-1 microphones as non-reference microphones, and estimate the time delay of the wake-up word speech segment between the non-reference microphone and the reference microphone ,in , It is The time delay of the wake-up word speech segment of the non-reference microphone relative to the reference microphone; based on the geometric layout of the microphone array, construct M-1 equations: ; in It is The position vectors of the non-reference microphones relative to the reference microphone, is the speed of sound, is the horizontal azimuth of the sound source, is the pitch angle of the sound source; Step S3230: Solve the M-1 equations simultaneously to obtain the horizontal azimuth angle of the wake-up word speaker relative to the vehicle microphone array. and distance ,in .

[0075] Specifically, when a custom wake-up word is detected, the microphone array starts to synchronously collect the wake-up word audio data, including M-channel voice signals , corresponding to the received signal of the wake-up word on M microphones. Through speech endpoint detection and cross-correlation analysis, the time boundary of the wake-up word speech segment can be estimated , and the time delay of the wake-up word speech segment between the non-reference microphone and the reference microphone .

[0076] Time Delay Reflects the wake-up word voice reaching the The time difference between the non-reference microphone and the reference microphone contains the azimuth of the sound source. and distance Through the simultaneous equations, we can get the azimuth of the sound source and pitch angle Further, the estimated azimuth Substituting into the far-field model of the microphone array, the distance of the sound source can be obtained The far-field model of microphone arrays is common knowledge in the field of acoustics and signal processing and is widely used in sound source localization and audio signal processing.

[0077] For example: Assume that the vehicle microphone array contains 4 microphone units , located in the front, back, left and right of the car, forming a rectangular array. When the wake-up word "Hello, Xiao X" is detected, the four microphones synchronously collect 100ms of audio data, and the signal received by reference microphone 1 is recorded as , the signal received by non-reference microphone 2 is recorded as , and so on. The speed of sound is generally 340m / s. By performing cross-correlation operations on these four voice signals, the following time delays are estimated: ; Substituting into the equation system: ; ; The known microphone position vector is: ; Solving the equations together yields: ; The estimated azimuth Substituting into the far field model we get the distance estimate: ;

[0078] Therefore, the speaker of the wake-up word is located at 60° in the front right of the car, about 0.4m away from the microphone array. This indicates that the wake-up word comes from the passenger area in the car, and the next step of voiceprint verification can be continued. If the positioning result shows that the wake-up word comes from a farther position outside the car, it can be determined as a false wake-up caused by environmental noise, and the wake-up request should be rejected.

[0079] Step S3300: Setting the wake-up angle threshold range and distance threshold range ;like and , it is determined that the detected custom wake-up word comes from a reasonable position in the car and is identified as a wake-up word to be confirmed. Otherwise, it is regarded as a false wake-up caused by noise outside the car, and the wake-up is rejected and the process jumps to step S3100 to continue the next round of in-car audio data collection and custom wake-up word detection; Furthermore, if Figure 5 As shown, step S3300 includes: Step S3310, establishing a three-dimensional space model of the interior of the vehicle according to the spatial structure parameters and seat layout of the vehicle model; calibrating the installation position of the vehicle microphone array in the three-dimensional space model of the interior of the vehicle; Step S3320: In the three-dimensional space model inside the vehicle, define wake-up areas, each of which is represented by a spatial polygon, and the coordinates of the corner points of the polygon are recorded. is a positive integer; Step S3330: The wake-up areas are projected onto a horizontal plane with the microphone array as the origin, and the azimuth span of each projection polygon is extracted. and distance span ,in is the total number of wake-up areas, is the minimum azimuth angle of the projected polygon, indicating the leftmost boundary of the wake-up area from the microphone array on the horizontal plane; is the maximum azimuth angle of the projected polygon, indicating the rightmost boundary of the wake-up area from the microphone array on the horizontal plane; is the shortest distance of the projected polygon, indicating the nearest boundary of the wake-up area starting from the microphone array; is the longest distance of the projected polygon, indicating the farthest boundary of the wake-up area from the microphone array; Step S3340: Take the union of all azimuth angle spans to determine the wake-up angle threshold range , ; Take the union of each distance span to determine the wake-up distance threshold range .

[0080] Specifically, step S3300 introduces the in-car layout and usage scenario priors on the basis of sound source positioning, sets the rationality requirements for the position of the wake-up word speaker, and further improves the reliability of wake-up confirmation. By modeling the three-dimensional space in the car and characterizing the position distribution characteristics of the driver and passengers, a three-dimensional passenger position template can be obtained. The speaker position obtained by microphone array positioning is mapped to the in-car space and compared with the passenger position template to determine whether the position is within a reasonable wake-up range. On this basis, the distance and angle threshold conditions are set to form a three-dimensional wake-up permission area. Only when the positioning result falls within the permitted area is the wake-up word considered to come from a legitimate user in the car, otherwise it is considered to be a false wake-up caused by noise interference outside the car. This method makes full use of the in-car scene prior information, sets the wake-up restriction conditions from the perspective of the spatial position of the sound source, and can effectively distinguish the outside noise wake-up from the real wake-up in the car, reducing the occurrence of false wake-ups. Compared with the traditional wake-up method, this step can more comprehensively describe the spatial characteristics of in-car voice interaction, and dynamically adjust the wake-up strategy using acoustic space information, making the in-car voice wake-up more flexible and accurate.

[0081] Set the wake-up angle threshold range and distance threshold range The purpose is to determine whether the sound source detected by the wake-up word comes from a reasonable position in the car based on the direction and distance of the sound source. The reasonable position here refers to the effective receiving range of the car microphone, usually the driver and co-pilot seat area. represents the horizontal angle of the sound source relative to the microphone array, and are the lower and upper limits of the wake-up angle in the car, such as 60° to 120°, which means that it only responds to sounds within a certain range in front of it; d represents the distance from the sound source to the microphone array, and The lower and upper limits of the wake-up distance in the car, such as 0.5 meters to 1.5 meters, indicate that the system will only respond to sounds within a certain distance.

[0082] When the detected wake-up word meets ∈ And d∈ , it can be preliminarily determined that it is a valid wake-up issued by the user in the car, and its voiceprint needs to be further verified; otherwise, it may be a false wake-up caused by users in other locations, audio of other devices, or noise outside the car, and it should be rejected directly to avoid unnecessary voiceprint comparison. For example, if the sound source of the detected wake-up word "Hello, Xiao X" is located in the front passenger seat, the horizontal angle with the microphone is 100°, and the distance is 1 meter, which belongs to the preset wake-up angle range [60°, 120°] and distance range [0.5m, 1.5m], it is temporarily determined to be a wake-up word to be confirmed, and the voiceprint verification process needs to be entered; if the sound source of the wake-up word "Hello, Xiao X" is located in the back seat, the horizontal angle with the microphone is 150° or the distance is 2 meters, it exceeds the preset angle or distance range, and it can be determined as a false wake-up. The wake-up should be rejected and the next round of wake-up word detection should be restarted.

[0083] This wake-up position judgment based on angle and distance makes full use of the beamforming capability of the microphone array, which can effectively reduce false wake-ups caused by sounds in non-target areas such as the rear seats and outside the car, and improve the system's ability to resist noise interference. It is a pre-screening for voiceprint verification, which can reduce unnecessary voiceprint comparisons, reduce the system's energy consumption and response delay. At the same time, the angle threshold and distance threshold can be flexibly set according to factors such as vehicle model and seat layout, and have a certain degree of adaptability.

[0084] Step S3400, extract acoustic features from the wake-up word audio data determined to be the wake-up word to be confirmed, and construct a second voiceprint feature; calculate the similarity score SIM between the second voiceprint feature and the first voiceprint feature. If the similarity score SIM is greater than the preset voiceprint verification threshold, the wake-up word to be confirmed is identified as a valid wake-up word, triggering the car screen to wake up; otherwise, it is identified as an invalid wake-up word, rejects the car screen wake-up and jumps to step S3100 to continue the next round of in-car audio data collection and custom wake-up word detection.

[0085] Specifically, the second voiceprint feature adopts the same feature extraction method as that used to construct the voiceprint feature library in step S1000, namely, fundamental frequency, resonance peak, speech rate, etc., to ensure the consistency of voiceprint verification. Then, the similarity score SIM between the second voiceprint feature and the first voiceprint feature obtained in step S1300 (i.e., the custom wake-up word voiceprint feature registered by the user) is calculated. The similarity score can be calculated using commonly used feature matching metrics such as Euclidean distance and cosine similarity. If the similarity score SIM is greater than the preset voiceprint verification threshold, such as 0.8, it is determined that the speaker of the wake-up word to be confirmed matches the voiceprint of the registered user, and is identified as a valid wake-up word, triggering the wake-up of the car screen; otherwise, if SIM is less than or equal to the verification threshold, it is determined that the speaker does not match the voiceprint of the registered user, and is identified as an invalid wake-up word of others, and the car screen wake-up is rejected.

[0086] It is worth mentioning that the setting of the voiceprint verification threshold needs to balance security and convenience. The higher the threshold, the stricter the voiceprint verification and the lower the false wake-up rate, but it may increase the rejection rate of my wake-up; the lower the threshold, the easier it is to pass the voiceprint verification, and it is more convenient for me to wake up, but the false wake-up rate may increase. The threshold can be flexibly adjusted according to the actual application scenario and user preferences. At the same time, as the number of times the user uses the custom wake-up word increases, the first voiceprint feature can be dynamically updated to improve the accuracy of voiceprint matching.

[0087] In short, steps S3300 and S3400 form two lines of defense through wake-up position judgment and voiceprint verification, which can effectively reduce the false wake-up rate of custom wake-up words and improve the security and reliability of vehicle-machine voice interaction. Compared with the traditional fixed wake-up word solution, this solution supports users to independently register personalized wake-up words, providing a better user experience; compared with the simple voiceprint recognition solution, this method integrates sound source positioning technology, introduces wake-up position judgment before voiceprint verification, reduces unnecessary voiceprint comparisons, and reduces system overhead. Therefore, while improving user convenience, this solution takes into account the real-time, accuracy and security of the system, providing a new idea for in-vehicle voice interaction.

[0088] Example 2 This embodiment provides a vehicle screen custom wake-up word configuration system based on voice control on the basis of embodiment 1, such as Figure 6 As shown, including: Noise level classification module: used to collect the in-vehicle noise data of the target vehicle and build an in-vehicle noise database, wherein the in-vehicle noise database includes n1 noise records; and classify the n1 noise records in the in-vehicle noise database into N1 noise levels; The first voiceprint feature acquisition module is used to obtain the user's custom wake-up word voice sample, extract the voiceprint feature vector of the custom wake-up word voice sample, and mark it as the custom wake-up word voiceprint feature vector; measure the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in the pre-built voiceprint feature library, and match to obtain the first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature vector; Model building module: used to build a wake-up word detection sub-model corresponding to N1 noise levels based on the in-vehicle noise database and the first voiceprint feature; Wake-up word recognition module: used to collect in-car audio data in real time, perform wake-up word recognition based on in-car audio data, wake-up word detection sub-model and first voiceprint features, and trigger the car screen to wake up.

[0089] In the noise level classification module, each of the n1 noise records includes attribute data of a noise segment; the attribute data of the noise segment includes a noise segment number, noise segment data, noise type label, noise energy value and noise spectrum feature vector; In the noise level classification module, the classification of n1 noise records in the in-vehicle noise database into N1 noise levels includes: Step S1210, calculating the energy mean E according to the noise energy value and noise spectrum feature vector of each noise record in the vehicle noise database i , spectrum center F i , and the spectral dispersion D i , which constitutes the noise acoustic feature vector [E i ,F i ,D i ]; among them, E i is the energy mean of the ith noise record, F i is the spectral center of the ith noise record, D i is the spectral dispersion of the ith noise record, [E i ,F i ,D i ] represents the noise acoustic feature vector of the i-th noise record; Step S1220, using the noise acoustic feature vector [E i ,F i ,D i ] is the feature description, clustering n1 noise records to obtain N1 noise clustering centers; Step S1230, calculate the average silhouette coefficient SC, if SC≤SC', adjust the number of noise cluster centers N1, return to step S1220 for re-clustering until SC>SC', and output the clustering result; SC' is a preset silhouette coefficient threshold.

[0090] In the first voiceprint feature acquisition module, obtaining the first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature vector includes: Step S1310, collecting voice samples of multiple users, extracting voiceprint feature vectors of the voice samples, and constructing a voiceprint feature library; the voiceprint feature vector includes fundamental frequency, formant and speech rate; Step S1320, measuring the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in the voiceprint feature library to obtain a similarity score; sorting according to the similarity score, selecting the voiceprint feature vector in the voiceprint feature library with the highest score, and marking it as the first voiceprint feature.

[0091] In the model construction module, the construction of the wake-up word detection sub-model corresponding to N1 noise levels includes: Step S2100, based on N1 noise levels in the in-vehicle noise database, construct N1 noise data subsets, each noise data subset corresponding to a noise level; The constructing of N1 noise data subsets includes: Traverse the n1 noise records in the in-vehicle noise database, and divide each noise record into corresponding N1 noise data subsets according to the noise level label of each noise record; Step S2200, constructing an independent wake-up word detection sub-model for each noise level; Step S2300, traverse N1 noise levels, execute step S2200, and obtain the wake-up word detection sub-model corresponding to all N1 noise levels. The output of the wake-up word detection sub-model is a custom wake-up word detection result represented by a Boolean value. If it is 1, it means that the custom wake-up word is detected. If it is 0, it means that the custom wake-up word is not detected.

[0092] The step S2200 includes: Step S2210, dividing the jth noise data subset into a training set, a validation set, and a test set according to a preset data set division ratio; 1≤j≤N1; Step S2220, fusing the training set in the j-th noise data subset with the first voiceprint feature to construct a wake-up word detection training set for noise level j; Step S2230, taking the wake-up word detection training set for noise level j as input, training the initial wake-up word detection sub-model under noise level j; Step S2240, using the validation set in the j-th noise data subset, optimizing the model parameters to obtain the final wake-up word detection sub-model under the noise level j; Step S2250, using the test set in the j-th noise data subset, evaluates the performance indicators of the final wake-up word detection sub-model under the noise level j.

[0093] In the wake-up word recognition module, the wake-up word recognition based on the in-vehicle audio data, the wake-up word detection sub-model and the first voiceprint feature to trigger the vehicle computer screen wake-up includes: Step S3100, collecting in-car audio data in real time, determining the noise level to which the in-car audio data belongs, and selecting a wake-up word detection sub-model corresponding to the noise level to which the in-car audio data belongs to perform custom wake-up word detection, and determining whether the custom wake-up word is detected; Step S3200: If no custom wake-up word is detected, jump to step S3100 to continue the next round of in-vehicle audio data collection and detection; if a custom wake-up word is detected, the wake-up word audio data is collected through the microphone array, the wake-up word speaker is located, and the horizontal azimuth angle of the wake-up word speaker relative to the vehicle microphone array is obtained. and distance ; Step S3300: Setting the wake-up angle threshold range and distance threshold range ;like and , it is determined that the detected custom wake-up word comes from a reasonable position in the car and is identified as a wake-up word to be confirmed. Otherwise, it is regarded as a false wake-up caused by noise outside the car, and the wake-up is rejected and the process jumps to step S3100 to continue the next round of in-car audio data collection and custom wake-up word detection; Step S3400, extract acoustic features from the wake-up word audio data determined to be the wake-up word to be confirmed, and construct a second voiceprint feature; calculate the similarity score SIM between the second voiceprint feature and the first voiceprint feature. If the similarity score SIM is greater than the preset voiceprint verification threshold, the wake-up word to be confirmed is identified as a valid wake-up word, triggering the car screen to wake up; otherwise, it is identified as an invalid wake-up word, rejects the car screen wake-up and jumps to step S3100 to continue the next round of in-car audio data collection and custom wake-up word detection.

[0094] The step S3100 includes: Step S3110, extracting features from the in-car audio data collected in real time, obtaining a real-time energy value Es, a real-time spectrum center of gravity Fs and a real-time spectrum dispersion Ds, and constructing a real-time acoustic feature vector [Es, Fs, Ds]; Step S3120, comparing the real-time acoustic feature vector [Es, Fs, Ds] with N1 noise cluster centers, calculating the Euclidean distance between the real-time acoustic feature vector and each noise cluster center, and selecting the noise level corresponding to the noise cluster center with the smallest Euclidean distance as the noise level to which the real-time collected in-vehicle audio data belongs; Step S3130: According to the noise level of the in-car audio data, a corresponding wake-up word detection sub-model is selected to perform customized wake-up word detection.

[0095] The step S3200 includes: Step S3210: If a custom wake-up word is detected, the wake-up word audio data is collected through the microphone array. The wake-up word audio data has a total of M voice signals, and the wake-up word audio data of the mth voice signal is recorded as , t is time, ; Step S3220: wake-up word audio data Perform voice endpoint detection and extract the voice received by each microphone. Find the start and end time of the wake-up word voice segment in , the wake-up word speech segment is recorded as ; Select one microphone from the microphone array as the reference microphone, and the other M-1 microphones as non-reference microphones, and estimate the time delay of the wake-up word speech segment between the non-reference microphone and the reference microphone ,in , It is The time delay of the wake-up word speech segment of the non-reference microphone relative to the reference microphone; based on the geometric layout of the microphone array, construct M-1 equations: ;

[0096] in It is The position vectors of the non-reference microphones relative to the reference microphone, is the speed of sound, is the horizontal azimuth of the sound source, is the pitch angle of the sound source; Step S3230: Solve the M-1 equations simultaneously to obtain the horizontal azimuth angle of the wake-up word speaker relative to the vehicle microphone array. and distance ,in .

[0097] The step S3300 includes: Step S3310, establishing a three-dimensional space model of the interior of the vehicle according to the spatial structure parameters and seat layout of the vehicle model; calibrating the installation position of the vehicle microphone array in the three-dimensional space model of the interior of the vehicle; Step S3320: In the three-dimensional space model inside the vehicle, define wake-up areas, each of which is represented by a spatial polygon, and the coordinates of the corner points of the polygon are recorded. is a positive integer; Step S3330: The wake-up areas are projected onto a horizontal plane with the microphone array as the origin, and the azimuth span of each projection polygon is extracted. and distance span ,in is the total number of wake-up areas, is the minimum azimuth angle of the projected polygon, indicating the leftmost boundary of the wake-up area from the microphone array on the horizontal plane; is the maximum azimuth angle of the projected polygon, indicating the rightmost boundary of the wake-up area from the microphone array on the horizontal plane; is the shortest distance of the projected polygon, indicating the nearest boundary of the wake-up area starting from the microphone array; is the longest distance of the projected polygon, indicating the farthest boundary of the wake-up area from the microphone array; Step S3340: Take the union of all azimuth angle spans to determine the wake-up angle threshold range , ; Take the union of each distance span to determine the wake-up distance threshold range .

[0098] Example 3 This embodiment discloses an electronic device, which may include one or more processors and one or more memories. The memories store computer-readable codes, which, when executed by the one or more processors, may execute the above-described method for configuring a custom wake-up word for a vehicle screen based on voice control.

[0099] The method or system according to the implementation of the present application can also be implemented with the help of the architecture of an electronic device. The electronic device may include a bus, one or more CPUs, a read-only memory (ROM), a random access memory (RAM), a communication port connected to a network, an input / output component, a hard disk, etc. A storage device in an electronic device, such as a ROM or a hard disk, can store the method for configuring a custom wake-up word for a vehicle screen based on voice control provided in the present application. The method for configuring a custom wake-up word for a car screen based on voice control may, for example, include: collecting in-car noise data of a target vehicle and constructing an in-car noise database, wherein the in-car noise database includes n1 noise records; dividing the n1 noise records in the in-car noise database into N1 noise levels, and recording the noise level label of each noise record; obtaining a user's custom wake-up word voice sample, extracting the voiceprint feature vector of the custom wake-up word voice sample, and marking it as a custom wake-up word voiceprint feature vector; measuring the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in a pre-constructed voiceprint feature library, and matching to obtain a first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature vector; constructing a wake-up word detection sub-model corresponding to N1 noise levels based on the in-car noise database and the first voiceprint feature; collecting in-car audio data in real time, performing wake-up word recognition based on the in-car audio data, the wake-up word detection sub-model and the first voiceprint feature, and triggering the car screen to wake up.

[0100] Furthermore, the electronic device may also include a user interface. Of course, the architecture disclosed in the present invention is only exemplary, and when implementing different devices, one or more components in the electronic device disclosed in the present invention may be omitted according to actual needs.

[0101] Example 4 This embodiment discloses a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the method for configuring a custom wake-up word for a vehicle screen based on voice control according to an embodiment of the present application can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache). Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0102] In addition, according to an embodiment of the present application, the process described above with reference to the flowchart may be implemented as a computer software program. For example, the present application provides a non-temporary machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application, for example: collecting in-car noise data of the target vehicle and building an in-car noise database, wherein the in-car noise database includes n1 noise records; dividing the n1 noise records in the in-car noise database into N1 noise levels, and recording the noise level label of each noise record; obtaining a user's custom wake-up word voice sample, extracting the voiceprint feature vector of the custom wake-up word voice sample, and marking it as a custom wake-up word voiceprint feature vector; measuring the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in a pre-built voiceprint feature library, and matching to obtain a first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature vector; constructing a wake-up word detection sub-model corresponding to N1 noise levels based on the in-car noise database and the first voiceprint feature; collecting in-car audio data in real time, performing wake-up word recognition based on the in-car audio data, the wake-up word detection sub-model and the first voiceprint feature, and triggering the vehicle computer screen to wake up. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are performed.

[0103] The methods, systems, and devices of the present application may be implemented in many ways. For example, the methods, systems, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.

[0104] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0105] The specific implementation modes described above further describe the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for configuring a custom wake-up word for a vehicle screen based on voice control, characterized in that: The method comprises: Collect the in-vehicle noise data of the target vehicle and build an in-vehicle noise database, wherein the in-vehicle noise database includes n1 noise records; divide the n1 noise records in the in-vehicle noise database into N1 noise levels, and record the noise level label of each noise record; obtain the user's custom wake-up word voice sample, extract the voiceprint feature vector of the custom wake-up word voice sample, and mark it as the custom wake-up word voiceprint feature vector; measure the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in the pre-built voiceprint feature library, and match the first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature vector; According to the in-car noise database and the first voiceprint feature, a wake-up word detection sub-model corresponding to N1 noise levels is constructed; Collect in-car audio data in real time, recognize the wake-up word based on the in-car audio data, the wake-up word detection sub-model and the first voiceprint feature, and trigger the car screen to wake up.

2. The method for configuring a custom wake-up word for a vehicle screen based on voice control according to claim 1, characterized in that: The n1 noise records, each noise record includes attribute data of a noise segment; the attribute data of the noise segment includes a noise segment number, noise segment data, noise type label, noise energy value and noise spectrum feature vector; The dividing the n1 noise records in the in-vehicle noise database into N1 noise levels includes: clustering the n1 noise records according to the noise energy values ​​and noise spectrum feature vectors of the n1 noise records in the in-vehicle noise database.

3. The method for configuring a custom wake-up word for a vehicle screen based on voice control according to claim 2, characterized in that: The clustering of n1 noise records includes: Step S1210, calculating the energy mean E according to the noise energy value and noise spectrum feature vector of each noise record in the vehicle noise database i , spectrum center F i , and the spectral dispersion D i , which constitutes the noise acoustic feature vector [E i ,F i ,D i ]; among them, E i is the energy mean of the ith noise record, F i is the spectral center of the ith noise record, D i is the spectral dispersion of the ith noise record, [E i ,F i ,D i ] represents the noise acoustic feature vector of the i-th noise record; Step S1220, using the noise acoustic feature vector [E i ,F i ,D i ] is the feature description, clustering n1 noise records to obtain N1 noise clustering centers; Step S1230, calculate the average silhouette coefficient SC, if SC≤SC', adjust the number of noise cluster centers N1, return to step S1220 for re-clustering until SC>SC', and output the clustering result; SC' is a preset silhouette coefficient threshold.

4. The method for configuring a custom wake-up word for a vehicle screen based on voice control according to claim 1, characterized in that: The method for constructing the voiceprint feature library is as follows: collecting voice samples of multiple users, extracting voiceprint feature vectors of the voice samples, and constructing the voiceprint feature library; The voiceprint feature vector includes fundamental frequency, formant and speech rate; The first voiceprint feature having the highest similarity to the custom wake-up word voiceprint feature vector comprises: Measure the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in the voiceprint feature library to obtain a similarity score; According to the similarity score sorting, the voiceprint feature vector in the voiceprint feature library with the highest score is selected and marked as the first voiceprint feature.

5. The method for configuring a custom wake-up word for a vehicle screen based on voice control according to claim 1, characterized in that: The construction of the wake-up word detection sub-model corresponding to N1 noise levels includes: Based on the N1 noise levels in the in-vehicle noise database, N1 noise data subsets are constructed, each noise data subset corresponds to a noise level; Traverse N1 noise levels, and build an independent wake-up word detection sub-model for each noise level; obtain the wake-up word detection sub-model corresponding to all N1 noise levels, and the output of the wake-up word detection sub-model is a custom wake-up word detection result represented by a Boolean value, if it is 1, it means that the custom wake-up word is detected, if it is 0, it means that the custom wake-up word is not detected; The constructing of N1 noise data subsets includes: Traverse the n1 noise records in the in-vehicle noise database, and divide each noise record into corresponding N1 noise data subsets according to the noise level label.

6. The method for configuring a custom wake-up word for a vehicle screen based on voice control according to claim 5, characterized in that: For each noise level, constructing an independent wake-up word detection sub-model includes: According to the preset data set division ratio, the jth noise data subset is divided into a training set, a validation set, and a test set; 1≤j≤N1; The training set in the j-th noise data subset is merged with the first voiceprint feature to construct a wake-up word detection training set for noise level j; Taking the wake-up word detection training set for noise level j as input, train the initial wake-up word detection sub-model under noise level j.

7. The method for configuring a custom wake-up word for a vehicle screen based on voice control according to claim 1, characterized in that: The performing of wake-up word recognition based on the in-vehicle audio data, the wake-up word detection sub-model and the first voiceprint feature to trigger the vehicle computer screen wake-up includes: Step S3100, determining the noise level to which the in-car audio data belongs, and selecting a wake-up word detection sub-model corresponding to the noise level to which the in-car audio data belongs to perform custom wake-up word detection, and determining whether the custom wake-up word is detected; Step S3200: If no custom wake-up word is detected, jump to step S3100 to continue the next round of in-vehicle audio data collection and detection; if a custom wake-up word is detected, the wake-up word audio data is collected through the microphone array, the wake-up word speaker is located, and the horizontal azimuth angle of the wake-up word speaker relative to the vehicle microphone array is obtained. and distance ; Step S3300: Setting the wake-up angle threshold range and distance threshold range ;like and , it is determined that the detected custom wake-up word comes from a reasonable position in the car and is identified as a wake-up word to be confirmed; otherwise, it is considered as a false wake-up, and the process jumps to step S3100 to continue the next round of in-car audio data collection and custom wake-up word detection; Step S3400, extract acoustic features from the wake-up word audio data determined to be the wake-up word to be confirmed, and construct a second voiceprint feature; calculate the similarity score SIM between the second voiceprint feature and the first voiceprint feature. If the similarity score SIM is greater than the preset voiceprint verification threshold, the wake-up word to be confirmed is identified as a valid wake-up word, triggering the car screen to wake up; otherwise, it is identified as an invalid wake-up word, rejects the car screen wake-up and jumps to step S3100 to continue the next round of in-car audio data collection and custom wake-up word detection.

8. The method for configuring a custom wake-up word for a vehicle screen based on voice control according to claim 7, characterized in that: The step of determining the noise level to which the in-vehicle audio data belongs includes: Extract features from the real-time collected in-car audio data to obtain the real-time energy value Es, the real-time spectrum center of gravity Fs and the real-time spectrum dispersion Ds, and construct the real-time acoustic feature vector [Es, Fs, Ds]; The real-time acoustic feature vector [Es, Fs, Ds] is compared with N1 noise cluster centers, the Euclidean distance between the real-time acoustic feature vector and each noise cluster center is calculated, and the noise level corresponding to the noise cluster center with the smallest Euclidean distance is selected as the noise level to which the real-time collected in-vehicle audio data belongs.

9. The method for configuring a custom wake-up word for a vehicle screen based on voice control according to claim 7, characterized in that: The horizontal azimuth angle of the wake-up word speaker relative to the vehicle microphone array is obtained and distance include: The wake-up word audio data is collected by a microphone array. The wake-up word audio data has M voice signals in total. The wake-up word audio data of the mth voice signal is recorded as , t is time, ; Wake-up word audio data Perform voice endpoint detection and extract the voice received by each microphone. Find the start and end time of the wake-up word voice segment in , the wake-up word speech segment is recorded as ; Select one microphone from the microphone array as the reference microphone, and the other M-1 microphones as non-reference microphones, and estimate the time delay of the wake-up word speech segment between the non-reference microphone and the reference microphone ,in , It is The time delay of the wake-up word speech segment of the non-reference microphone relative to the reference microphone; Based on the geometric layout of the microphone array, construct M-1 equations: ； in It is The position vectors of the non-reference microphones relative to the reference microphone, is the speed of sound, is the horizontal azimuth of the sound source, is the pitch angle of the sound source; Solve the M-1 equations simultaneously to obtain the horizontal azimuth angle of the wake-up word speaker relative to the microphone array: and distance ,in .

10. A vehicle screen custom wake-up word configuration system based on voice control, which is used to implement a vehicle screen custom wake-up word configuration method based on voice control as claimed in any one of claims 1 to 9, characterized in that: The system comprises: Noise level classification module: used to collect the in-vehicle noise data of the target vehicle and build an in-vehicle noise database, wherein the in-vehicle noise database includes n1 noise records; and classify the n1 noise records in the in-vehicle noise database into N1 noise levels; The first voiceprint feature acquisition module is used to obtain the user's custom wake-up word voice sample, extract the voiceprint feature vector of the custom wake-up word voice sample, and mark it as the custom wake-up word voiceprint feature vector; measure the similarity between the custom wake-up word voiceprint feature vector and each voiceprint feature vector in the pre-built voiceprint feature library, and match to obtain the first voiceprint feature with the highest similarity to the custom wake-up word voiceprint feature vector; Model building module: used to build a wake-up word detection sub-model corresponding to N1 noise levels based on the in-vehicle noise database and the first voiceprint feature; Wake-up word recognition module: used to collect in-car audio data in real time, perform wake-up word recognition based on in-car audio data, wake-up word detection sub-model and first voiceprint features, and trigger the car screen to wake up.

Images