BCM module cooperative control method based on voice instruction recognition in vehicle-mounted high-noise environment

Abstract

The application relates to the technical field of artificial intelligence, and discloses a BCM module cooperative control method based on voice instruction recognition in a vehicle high-noise environment, which comprises the following steps: collecting original voice instruction data and vehicle state parameter data in the vehicle high-noise environment; performing voice signal preprocessing in the noise environment based on the original voice instruction data to generate denoised voice instruction data; through multi-level noise suppression processing combining adaptive filtering and spectral subtraction, engine noise, wind noise and steady-state background noise are filtered out, and residual noise is eliminated for sudden impact noise, so that the intelligibility of the voice signal is improved. Meanwhile, through a voice activity detection algorithm, and by dynamically adjusting the detection threshold according to the vehicle state parameters, effective voice segments and noise segments can be separated, the accuracy of voice feature extraction in a complex time-varying noise environment is ensured, and the robustness and reliability of the voice instruction recognition are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment. Background Technology

[0002] Currently, due to the complex and time-varying nature of in-vehicle environmental noise, the noise reduction algorithms equipped for voice command recognition, while capable of suppressing steady-state noise, cannot fully adapt to the impact of non-steady noise caused by rapid engine acceleration or sudden road bumps. When the input voice command is contaminated by such sudden high-energy noise, it will cause serious distortion of voice features, leading to misjudgment by the subsequent recognition model and failing to guarantee the robustness of voice interaction in high-noise environments.

[0003] Therefore, a BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment is proposed to solve the above problems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a BCM module collaborative control method for voice command recognition in high-noise vehicle environments. This method solves the problem mentioned in the background that severe distortion of voice features leads to misjudgments in subsequent recognition models, making it impossible to guarantee the robustness of voice interaction in high-noise environments.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment, the method comprising the following steps:

[0006] S1. Collect raw voice command data and vehicle status parameter data in a high-noise vehicle environment;

[0007] S2. Based on the original voice command data, perform voice signal preprocessing under noisy conditions to generate denoised voice command data;

[0008] S3. Based on the denoised voice command data, perform voice command feature extraction processing to generate voice command feature vector data;

[0009] S4. Based on the voice command feature vector data, perform voice command recognition model matching processing to generate preliminary voice command recognition result data;

[0010] S5. Combine the vehicle state parameter data to perform BCM collaborative control strategy analysis and generate BCM control command candidate set data;

[0011] S6. Based on the preliminary voice command recognition result data and the BCM control command candidate set data, perform command consistency verification processing to generate optimized voice command recognition data;

[0012] S7. Perform BCM collaborative control command execution processing based on the optimized voice command recognition data to generate BCM control execution feedback data;

[0013] S8. Based on the BCM control execution feedback data, perform performance evaluation and adaptive adjustment to generate dynamic parameter update data;

[0014] S9. Utilize the dynamic parameter update data to optimize the speech recognition model and BCM control strategy in real time, and generate adaptive collaborative control model data.

[0015] Preferably, the step S1 of collecting raw voice command data and vehicle status parameter data in a high-noise in-vehicle environment includes the following steps:

[0016] S11. Collect raw voice command data in a high-noise environment inside the vehicle through an in-vehicle microphone array, where the high-noise environment includes engine noise, wind noise and road noise.

[0017] S12. Obtain vehicle status parameter data, including vehicle speed, engine speed, window status and air conditioning operating parameters, through the vehicle CAN bus;

[0018] S13. Synchronize the original voice command data and vehicle status parameter data with timestamps and store them in the vehicle embedded database.

[0019] Preferably, the preprocessing of speech signals in a noisy environment in step S2 includes the following steps:

[0020] S21. An adaptive filter is used to suppress background noise in the original speech command data to generate preliminary denoised speech data.

[0021] S22. Based on spectral subtraction, residual noise is eliminated from the preliminary denoised speech data to generate secondary denoised speech data;

[0022] S23. Extract effective speech segments through a speech activity detection algorithm and generate denoised speech command data, wherein the speech activity detection threshold is dynamically adjusted according to vehicle status parameters.

[0023] Preferably, the voice command feature extraction process in step S3 includes the following steps:

[0024] S31. The denoised voice command data is processed by framing and windowing, with a frame length of 25ms and a frame shift of 10ms.

[0025] S32. Extract the Mel frequency cepstral coefficient features and linear prediction coding features of each frame of speech to generate a basic feature vector;

[0026] S33. Combine contextual information with recurrent neural network to perform temporal feature fusion and generate voice command feature vector data.

[0027] Preferably, the voice command recognition model matching process in S4 includes the following steps:

[0028] S41. Establish a speech command recognition model library, including hidden Markov models, deep neural network models, and end-to-end sequence models;

[0029] S42. Input the voice command feature vector data into the model library for parallel recognition to generate multiple candidate recognition results with confidence scores;

[0030] S43. Based on the confidence score of the candidate recognition results and the vehicle state parameter data, select the optimal candidate and generate preliminary voice command recognition result data.

[0031] Preferably, the BCM collaborative control strategy analysis in step S5 includes the following steps:

[0032] S51. Determine the executable BCM control command type based on the vehicle speed and engine speed selected from the vehicle status parameter data.

[0033] S52. Within the range of executable BCM control instruction types, a candidate set of BCM control instructions is generated based on the instruction frequency in historical control data and the risk level in the security rules.

[0034] S53. Using a fuzzy logic algorithm and combining historical selection data from the user preference model, candidate instructions are prioritized to generate a BCM control instruction candidate set data.

[0035] Preferably, the instruction consistency verification process in S6 includes the following steps:

[0036] S61. Perform semantic matching analysis on the preliminary voice command recognition result data and the BCM control command candidate set data to generate a similarity evaluation result;

[0037] S62. Based on the similarity evaluation results, perform instruction consistency verification and generate verification conclusion data;

[0038] S63. When the verification result is consistent, directly output the optimized voice command recognition data; otherwise, trigger the re-recognition process based on the context information.

[0039] Preferably, the BCM collaborative control instruction execution process in S7 includes the following steps:

[0040] S71. Parse the optimized voice command recognition data into a control command sequence through the BCM communication interface;

[0041] S72. Execute the control command sequence and monitor the changes in the execution status of the BCM control commands in real time, and generate status log data;

[0042] S73. When the status is abnormal, enable the backup control mechanism and generate BCM control execution feedback data.

[0043] Preferably, the performance evaluation and adaptive adjustment in S8 includes the following steps:

[0044] S81. Based on the BCM control execution feedback data, analyze and generate a performance evaluation report, which includes recognition accuracy and response efficiency indicators.

[0045] S82. Based on the performance evaluation report, dynamically optimize the internal parameters and environmental noise adaptation threshold of the speech recognition model;

[0046] S83. Based on the optimized parameters and thresholds, generate and upload dynamic parameter update data packets to the cloud server for iterative learning.

[0047] Preferably, the real-time optimization of the speech recognition model and BCM control strategy in S9 includes the following steps:

[0048] S91. The dynamic parameters are called to update the data packet, and the speech recognition model is incrementally trained and optimized through machine learning algorithms.

[0049] S92. Place the optimized model in a virtual BCM control scenario for simulation verification and generate strategy effectiveness verification results;

[0050] S93. Integrate the verification results with the current model data, update and output the adaptive cooperative control model data.

[0051] Compared with existing technologies, this invention provides a BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment, which has the following advantages:

[0052] 1. In this invention, when performing voice command recognition in a high-noise vehicle environment, a multi-level noise suppression process combining adaptive filtering and spectral subtraction is employed to filter out engine noise, wind noise, and steady-state background noise. Residual noise is eliminated for sudden impact noise, thus improving the clarity of the voice signal. Simultaneously, a voice activity detection algorithm is used, with its detection threshold dynamically adjusted according to vehicle state parameters. This allows for the separation of effective voice segments from noise segments, ensuring the accuracy of voice feature extraction in complex time-varying noise environments. This improves the robustness and reliability of voice command recognition, avoiding feature distortion and misrecognition problems caused by noise pollution.

[0053] 2. In this invention, when mapping and making decisions from voice commands to BCM control commands, a context-aware command decision-making mechanism is constructed by real-time collection and deep fusion of vehicle speed, engine speed, and vehicle state parameters. The confidence level of voice recognition is analyzed collaboratively with the real-time vehicle driving status, and the control commands are risk-assessed and filtered according to preset safety rules. This enables intelligent judgment of the risk level of the current driving scenario and proactive avoidance of non-critical control commands that may distract the driver in high-risk conditions, thereby enhancing the proactive safety and situational adaptability of the voice control system.

[0054] 3. In this invention, during the collaborative process of speech recognition and BCM control execution, an instruction consistency verification mechanism is introduced to perform semantic matching verification between the preliminary recognition results and the executable control instruction set. This can intercept and process semantically ambiguous and unambiguous instructions. When verification fails or control execution is abnormal, it can trigger re-recognition and enable the backup control mechanism, thereby improving the accuracy and stability of speech recognition and control execution and enhancing the ability to cope with abnormal situations. Attached Figure Description

[0055] Figure 1 This is a flowchart of the BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment according to the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] For specific implementation examples, please refer to: Figure 1 A collaborative control method for BCM modules based on voice command recognition in high-noise vehicle environments, comprising the following steps:

[0058] S1. Collect raw voice command data and vehicle status parameter data in a high-noise vehicle environment;

[0059] S2. Based on the original voice command data, perform voice signal preprocessing in a noisy environment to generate denoised voice command data;

[0060] S3. Based on the denoised speech command data, perform speech command feature extraction processing to generate speech command feature vector data;

[0061] S4. Based on the voice command feature vector data, perform voice command recognition model matching processing to generate preliminary voice command recognition result data;

[0062] S5. Combine vehicle status parameter data to perform BCM collaborative control strategy analysis and generate BCM control command candidate set data;

[0063] S6. Based on the preliminary voice command recognition result data and the BCM control command candidate set data, perform command consistency verification processing to generate optimized voice command recognition data;

[0064] S7. Process BCM collaborative control command execution based on optimized voice command recognition data, and generate BCM control execution feedback data;

[0065] S8. Based on the BCM control execution feedback data, perform performance evaluation and adaptive adjustment, and generate dynamic parameter update data;

[0066] S9. Utilize dynamic parameter update data to optimize the speech recognition model and BCM control strategy in real time, generating adaptive collaborative control model data.

[0067] The steps involved in collecting raw voice command data and vehicle status parameter data in a high-noise in-vehicle environment in S1 are as follows:

[0068] S11. Collect raw voice command data in a high-noise environment inside the vehicle through an in-vehicle microphone array, where the high-noise environment includes engine noise, wind noise and road noise.

[0069] S12. Obtain vehicle status parameter data, including vehicle speed, engine speed, window status and air conditioning operating parameters, through the vehicle CAN bus;

[0070] S13. Synchronize the original voice command data and vehicle status parameter data with timestamps and store them in the vehicle embedded database.

[0071] S2 performs speech signal preprocessing in noisy environments, including the following steps:

[0072] S21. An adaptive filter is used to suppress background noise in the original speech command data to generate preliminary denoised speech data, including the following steps:

[0073] S211. Separate the reference noise signal, which is mainly composed of engine noise, from the original voice command data;

[0074] S212. The least mean square algorithm is used as the core algorithm of the adaptive filter. The filter coefficients are dynamically updated based on the correlation between the reference noise signal and the original speech signal. The update process of the filter coefficients is defined by the following formula:

[0075] ;

[0076] ;

[0077] in, for The filter coefficient vector at time t. for The waveguide coefficient vector at time 1. The convergence factor is for Error signal at time, for The expected response signal at time. for The input signal vector at time t;

[0078] S22. Based on spectral subtraction, residual noise is eliminated from the initially denoised speech data to generate secondary denoised speech data, including the following steps:

[0079] S221. Perform a short-time Fourier transform on the initially denoised speech data to convert it from the time domain to the frequency domain, obtaining the speech spectrum, including the following sub-steps:

[0080] The initial denoised speech data is processed by frame segmentation, with a frame length of N points and a frame shift of R points.

[0081] Each frame of speech data is multiplied by a window function to reduce spectral leakage;

[0082] A Fast Fourier Transform (FFT) is performed on each windowed frame of data to transform it from the time domain to the frequency domain; the Short Time Fourier Transform (SFT) is defined by the following formula:

[0083] ;

[0084] in, For the first Frame speech in the Complex spectral values ​​at each frequency point It is a time-domain speech signal. For window functions, For frame length, For frame shift, Dummy unit For sampling point index;

[0085] S222. Infer the noise power spectrum in the speech spectrum and subtract the noise power spectrum from the speech spectrum to obtain the enhanced spectrum; the core of this spectrum subtraction operation is defined by the following formula:

[0086] ;

[0087] in, For the first Frame number Power spectrum enhanced at each frequency point The power spectrum of noisy speech. The inferred noise power spectrum, For over-subtraction factor, The lower limit factor of the spectrum;

[0088] S223. Perform a short-time inverse Fourier transform on the enhanced spectrum to restore it from the frequency domain to the time domain, generating secondary denoised speech data;

[0089] S23. Extract effective speech segments using a speech activity detection algorithm and generate denoised speech command data, wherein the speech activity detection threshold is dynamically adjusted according to vehicle status parameters, including the following steps:

[0090] S231. Calculate the short-time energy and short-time zero-crossing rate of the secondary denoised speech data; wherein, the calculation of short-time energy and short-time zero-crossing rate is defined by the following formulas:

[0091] ;

[0092] ;

[0093] in, The short-time energy of a single frame of speech. For the first time in this frame The amplitude of each sampling point The number of sampling points in one frame. For short-time zero crossing rate, For sampling point index;

[0094] S232. Compare the short-time energy with a first threshold dynamically adjusted according to the current vehicle state parameters, and compare the short-time zero-crossing rate with a preset second threshold.

[0095] S233. When the short-time energy is higher than the first threshold and the short-time zero-crossing rate is lower than the second threshold, the current frame is determined to be a valid speech frame; otherwise, it is determined to be a noise frame.

[0096] S234. Merge consecutive valid speech frames to generate a valid speech segment.

[0097] The voice command feature extraction process in S3 includes the following steps:

[0098] S31. Perform frame segmentation and windowing processing on the denoised voice command data, with a frame length of 25ms and a frame shift of 10ms.

[0099] S32. Extract the Mel frequency cepstral coefficient features and linear prediction coding features of each frame of speech to generate a basic feature vector, which includes the following sub-steps:

[0100] S321. Perform pre-emphasis processing on each frame of speech signal to enhance high-frequency components and compensate for the high-frequency parts of the speech signal suppressed by the vocal system; wherein, the pre-emphasis process is defined by the following formula:

[0101]

[0102] in, For the first time after pre-aggravation Each sampling point signal, For the original number Each sampling point signal, This is the pre-emphasis coefficient;

[0103] S322. Perform a fast Fourier transform on the pre-emphasized speech signal to convert the time-domain signal into a frequency-domain spectrum.

[0104] S323. Pass the frequency domain spectrum through a Mel-scale filter bank, calculate the logarithmic energy of the output of each filter, and obtain the Mel spectrum:

[0105] ;

[0106] in, For the first The logarithmic energy of the output of a Mel filter. For the first Frame speech in the Spectral amplitude at each frequency point For the first The Mel filter in the first... Frequency response at each frequency point;

[0107] S324. Perform a discrete cosine transform on the Mel frequency spectrum and take the first 12 coefficients as the characteristics of the Mel frequency cepstral coefficients.

[0108] S325. Perform linear prediction analysis on each frame of speech signal using autocorrelation analysis to calculate the linear prediction coding coefficients:

[0109] ;

[0110] in, For linear prediction order, The first one to be solved Linear prediction coding coefficients, For delay The autocorrelation value;

[0111] S326. Combine the Mel frequency cepstral coefficient features with the linear prediction coding coefficients to generate a basic feature vector;

[0112] S33. Combine contextual information with a recurrent neural network to perform temporal feature fusion and generate speech command feature vector data, including the following steps:

[0113] S331. Input the basic feature vector sequence into a recurrent neural network with a long short-term memory network structure.

[0114] S332. The Long Short-Term Memory (LSTM) network processes the basic feature vectors step-by-step, and its hidden states retain the contextual information from previous time steps. The cell state update of the LTM network at time step t is defined by the following formula:

[0115] ;

[0116] in, For time step The cell state vector, For time step The forget gate activation vector, For time step The input gate activation vector, For time step The candidate cell state vector;

[0117] S333. Output the hidden state of the last time step as the voice command feature vector data that incorporates contextual information.

[0118] The voice command recognition model matching process in S4 includes the following steps:

[0119] S41. Establish a speech command recognition model library, including hidden Markov models, deep neural network models, and end-to-end sequence models;

[0120] S42. Input the voice command feature vector data into the model library for parallel recognition, and generate multiple candidate recognition results with confidence scores, including the following steps:

[0121] S421. Each model in the model library recognizes the input voice command feature vector data and outputs a probability distribution vector, where each element corresponds to the probability of a preset voice command.

[0122] S422. For each model, the highest probability value in its output probability distribution is taken as the confidence score of that model for that candidate recognition result. The confidence score is calculated by the following formula:

[0123] ;

[0124] in, Score the confidence level. This is the probability distribution vector output by the model. For the first element in this vector The probability value corresponding to each preset voice command;

[0125] S423. Summarize the candidate recognition results output by all models and their corresponding confidence scores;

[0126] S43. Based on the confidence score of the candidate recognition results and the vehicle state parameter data, select the optimal candidate and generate preliminary voice command recognition result data, including the following steps:

[0127] S431. Set weighting factors for confidence scores and vehicle status parameters respectively;

[0128] S432. For each candidate identification result, a comprehensive score is calculated based on its confidence score and the current vehicle state parameters, namely vehicle speed and engine speed; wherein, the comprehensive score is defined by the following formula:

[0129] ;

[0130] ;

[0131] in, For the overall score, Score the confidence level. Let vehicle speed be the influence function. Current vehicle speed Engine speed, The maximum safe speed for allowing voice operation, The function is the effect of engine speed. These are the weighting coefficients;

[0132] S433. Select the candidate recognition result with the highest comprehensive score as the optimal candidate.

[0133] The BCM collaborative control strategy analysis in S5 includes the following steps:

[0134] S51. Based on the vehicle speed and engine speed selected from the vehicle status parameter data, determine the type of executable BCM control command.

[0135] S52. Within the range of executable BCM control instruction types, generate a candidate set of BCM control instructions based on the instruction frequency in historical control data and the risk level in safety rules.

[0136] S53. Using a fuzzy logic algorithm and combining historical selection data from the user preference model, prioritize candidate instructions to generate a BCM control instruction candidate set, including the following steps:

[0137] S531. Define the fuzzy set and membership function for the input variables "instruction frequency", "risk level" and "user preference";

[0138] S532. Establish a fuzzy rule base, which includes rules that prioritize high-frequency instructions with low-risk levels.

[0139] S533. The centroid method is used to defuzzify the output of fuzzy inference, converting the fuzzy quantities into precise priority values; wherein, the centroid method in the defuzzification process is defined by the following formula:

[0140] ;

[0141] in, To obtain the precise priority value after defuzzification, To output the first in the universe of discourse One element, for membership degree This is to output the number of points after the universe of discourse is discretized.

[0142] S534. Sort the BCM control instruction candidate set according to the priority value.

[0143] The instruction consistency verification process in S6 includes the following steps:

[0144] S61. Perform semantic matching analysis on the preliminary voice command recognition result data and the BCM control command candidate set data to generate similarity evaluation results, including the following steps:

[0145] S611. Convert the preliminary voice command recognition result data and the command text in the BCM control command candidate set into word vectors;

[0146] S612. Calculate the cosine similarity between the word vectors of the preliminary speech command recognition result data and the word vectors of each candidate command; wherein, the calculation of cosine similarity is defined by the following formula:

[0147] ;

[0148] in, The cosine similarity value is... , For two word vectors, , For the vector of the first Each dimension component The dimension of the word vector;

[0149] S613. The calculated cosine similarity value is used as the similarity evaluation result;

[0150] S62. Based on the similarity assessment results, perform instruction consistency verification and generate verification conclusion data;

[0151] S63. When the verification result is consistent, directly output the optimized voice command recognition data; otherwise, trigger the re-recognition process based on the context information.

[0152] The BCM collaborative control instruction execution process in S7 includes the following steps:

[0153] S71. Optimize voice command recognition data into a sequence of control commands via the BCM communication interface;

[0154] S72. Execute the control command sequence and monitor the changes in the execution status of the BCM control commands in real time, and generate status log data;

[0155] S73. When the status is abnormal, enable the backup control mechanism and generate BCM control execution feedback data.

[0156] Performance evaluation and adaptive tuning in S8 includes the following steps:

[0157] S81. Based on the BCM control execution feedback data, analyze and generate a performance evaluation report, which includes recognition accuracy and response efficiency indicators.

[0158] S82. Based on the performance evaluation report, dynamically optimize the internal parameters and environmental noise adaptation threshold of the speech recognition model, including the following steps:

[0159] S821. When the recognition accuracy in the performance evaluation report is lower than the preset target, adjust the weight decay coefficient of the fully connected layer inside the model; the adjustment of the weight decay coefficient is defined by the following formula:

[0160] ;

[0161] in, This is the adjusted weight decay coefficient. This is the weight decay coefficient before adjustment. To adjust the rate parameters, For target recognition accuracy, This represents the current recognition accuracy.

[0162] S822. When the response efficiency in the performance evaluation report is lower than the preset target, the first threshold in the speech activity detection algorithm is adjusted proportionally according to the deviation between the response efficiency and the preset target. Specifically, the greater the deviation of the response efficiency from the preset target, the greater the adjustment of the first threshold in the direction of reducing speech activation sensitivity.

[0163] S823. Use the adjusted weight attenuation coefficient and the first threshold as the new internal parameters and noise adaptation threshold;

[0164] S83. Based on the optimized parameters and thresholds, generate and upload dynamic parameter update data packets to the cloud server for iterative learning.

[0165] Real-time optimization of the speech recognition model and BCM control strategy in S9 includes the following steps:

[0166] S91. Update the data packet using dynamic parameters, and incrementally train and optimize the speech recognition model using machine learning algorithms, including the following steps:

[0167] S911: Download the newly added speech training samples contained in the dynamic parameter update data package from the cloud server;

[0168] S912. Using stochastic gradient descent, the speech recognition model is trained on newly added speech training samples with the current parameters as initial values; the parameter update process of stochastic gradient descent is defined by the following formula:

[0169] ;

[0170] ;

[0171] in, For the first Model parameters for the next iteration For the first The learning rate for the next iteration. For loss function, , For training samples Inputs and outputs Indicates the parameter gradient, The initial learning rate, This is the learning rate decay coefficient;

[0172] S913. Save the model parameters obtained after training to complete incremental training and optimization;

[0173] S92. Place the optimized model in a virtual BCM control scenario for simulation verification and generate strategy effectiveness verification results;

[0174] S93. Integrate the verification results with the current model data, update and output the adaptive cooperative control model data.

[0175] The operation steps of the BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment are as follows:

[0176] Step 1: Synchronous Acquisition and Preprocessing of Multi-Source Data

[0177] This method first acquires raw voice command data in a high-noise environment using an onboard microphone array, and simultaneously obtains vehicle status parameter data, including vehicle speed and engine speed, in real time via the vehicle's CAN bus. The acquired multi-source data is then timestamped and stored in an onboard embedded database, providing complete and consistent input information for subsequent processing.

[0178] Step 2: Two-level noise suppression and speech enhancement:

[0179] To address the non-stationary nature of vehicle noise, an adaptive filter is employed to suppress background noise in the original speech, generating preliminary denoised speech data. Subsequently, spectral subtraction is used to further eliminate residual noise, addressing sudden noise spikes. Finally, a speech activity detection algorithm, combined with dynamically adjusted threshold parameters, is used to extract effective speech segments, generating high-quality denoised speech command data, laying the foundation for feature extraction.

[0180] Step 3: Robust speech feature extraction and temporal modeling:

[0181] After preprocessing the denoised speech data, including framing and windowing, the Mel-frequency cepstral coefficients and linear predictive coding features of each frame are extracted to construct a basic feature vector. Subsequently, the long short-term memory structure of a recurrent neural network is used to perform temporal fusion of the feature sequences to generate speech command feature vector data containing contextual information, thereby improving the robustness of the features to noise interference.

[0182] Step 4: Multi-model confidence decision and security strategy analysis:

[0183] Voice command feature vector data is input into a recognition model library composed of multiple models for parallel recognition, generating candidate recognition results with confidence scores. Based on the confidence scores and real-time vehicle state parameters, a weighted decision algorithm is used to calculate a comprehensive score, and the optimal candidate is selected as the initial voice command recognition result. Simultaneously, based on the vehicle state, a BCM (Balanced Vehicle Control) collaborative control strategy analysis is performed. A candidate set of BCM control commands is generated by combining historical data and safety rules, and the commands are prioritized to ensure the safety and rationality of the control commands.

[0184] Step 5: Instruction Consistency Verification and Secure Execution

[0185] Semantic matching analysis is performed on the initial voice command recognition results and the BCM control command candidate set. Cosine similarity calculation is used to generate a similarity evaluation result, completing the command consistency verification. Verified commands are parsed into control command sequences via the BCM communication interface and executed, with execution status monitored in real time. When an anomaly occurs or verification fails, a backup control mechanism is immediately activated and a re-recognition process is triggered, generating BCM control execution feedback data to ensure the reliability and security of the control process.

[0186] Step Six: Performance Evaluation and Closed-Loop Adaptive Optimization

[0187] Based on BCM control execution feedback data, performance evaluation reports including indicators such as recognition accuracy and response efficiency are generated periodically. Based on the evaluation results, the internal parameters of the speech recognition model and the speech activity detection threshold are dynamically optimized. The optimized parameters are uploaded to the cloud server as a dynamic parameter update data package for incremental training and optimization of the model. After verifying the optimization effect through virtual BCM control scenario simulation, the adaptive collaborative control model data is updated to achieve continuous self-improvement and closed-loop optimization of this method.

[0188] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0189] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment, characterized in that: The method includes the following steps: S1. Collect raw voice command data and vehicle status parameter data in a high-noise vehicle environment; S2. Based on the original voice command data, perform voice signal preprocessing under noisy conditions to generate denoised voice command data; S3. Based on the denoised voice command data, perform voice command feature extraction processing to generate voice command feature vector data; S4. Based on the voice command feature vector data, perform voice command recognition model matching processing to generate preliminary voice command recognition result data; S5. Analyze the BCM collaborative control strategy based on the vehicle state parameter data to generate a candidate set of BCM control commands, including the following steps: S51. Determine the executable BCM control command type based on the vehicle speed and engine speed selected from the vehicle status parameter data. S52. Within the range of executable BCM control instruction types, a candidate set of BCM control instructions is generated based on the instruction frequency in historical control data and the risk level in the security rules. S53. Using a fuzzy logic algorithm and combining historical selection data from the user preference model, prioritize the candidate instructions to generate a BCM control instruction candidate set data. S6. Based on the preliminary voice command recognition result data and the BCM control command candidate set data, perform command consistency verification processing to generate optimized voice command recognition data; S7. Perform BCM collaborative control command execution processing based on the optimized voice command recognition data to generate BCM control execution feedback data; S8. Based on the BCM control execution feedback data, perform performance evaluation and adaptive adjustment to generate dynamic parameter update data; S9. Utilize the dynamic parameter update data to optimize the speech recognition model and BCM control strategy in real time, and generate adaptive collaborative control model data.

2. The BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment according to claim 1, characterized in that: The steps in S1 for collecting raw voice command data and vehicle status parameter data in a high-noise in-vehicle environment include: S11. Collect raw voice command data in a high-noise environment inside the vehicle through an in-vehicle microphone array, where the high-noise environment includes engine noise, wind noise and road noise. S12. Obtain vehicle status parameter data, including vehicle speed, engine speed, window status and air conditioning operating parameters, through the vehicle CAN bus; S13. Synchronize the original voice command data and vehicle status parameter data with timestamps and store them in the vehicle embedded database.

3. The BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment according to claim 2, characterized in that: The preprocessing of speech signals in a noisy environment in step S2 includes the following steps: S21. An adaptive filter is used to suppress background noise in the original speech command data to generate preliminary denoised speech data. S22. Based on spectral subtraction, residual noise is eliminated from the preliminary denoised speech data to generate secondary denoised speech data; S23. Extract effective speech segments through a speech activity detection algorithm and generate denoised speech command data, wherein the speech activity detection threshold is dynamically adjusted according to vehicle status parameters.

4. The BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment according to claim 3, characterized in that: The voice command feature extraction process in S3 includes the following steps: S31. The denoised voice command data is processed by framing and windowing, with a frame length of 25ms and a frame shift of 10ms. S32. Extract the Mel frequency cepstral coefficient features and linear prediction coding features of each frame of speech to generate a basic feature vector; S33. Combine contextual information with recurrent neural network to perform temporal feature fusion and generate voice command feature vector data.

5. The BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment according to claim 4, characterized in that: The voice command recognition model matching process in S4 includes the following steps: S41. Establish a speech command recognition model library, including hidden Markov models, deep neural network models, and end-to-end sequence models; S42. Input the voice command feature vector data into the model library for parallel recognition to generate multiple candidate recognition results with confidence scores; S43. Based on the confidence score of the candidate recognition results and the vehicle state parameter data, select the optimal candidate and generate preliminary voice command recognition result data.

6. The BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment according to claim 1, characterized in that: The instruction consistency verification process in S6 includes the following steps: S61. Perform semantic matching analysis on the preliminary voice command recognition result data and the BCM control command candidate set data to generate a similarity evaluation result; S62. Based on the similarity evaluation results, perform instruction consistency verification and generate verification conclusion data; S63. When the verification result is consistent, directly output the optimized voice command recognition data; otherwise, trigger the re-recognition process based on the context information.

7. The BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment according to claim 6, characterized in that: The BCM collaborative control instruction execution process in S7 includes the following steps: S71. Parse the optimized voice command recognition data into a control command sequence through the BCM communication interface; S72. Execute the control command sequence and monitor the changes in the execution status of the BCM control commands in real time, and generate status log data; S73. When the status is abnormal, enable the backup control mechanism and generate BCM control execution feedback data.

8. The BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment according to claim 7, characterized in that: The performance evaluation and adaptive adjustment in S8 includes the following steps: S81. Based on the BCM control execution feedback data, analyze and generate a performance evaluation report, which includes recognition accuracy and response efficiency indicators. S82. Based on the performance evaluation report, dynamically optimize the internal parameters and environmental noise adaptation threshold of the speech recognition model; S83. Based on the optimized parameters and thresholds, generate and upload dynamic parameter update data packets to the cloud server for iterative learning.

9. The BCM module collaborative control method based on voice command recognition in a high-noise vehicle environment according to claim 8, characterized in that: The real-time optimization of the speech recognition model and BCM control strategy in S9 includes the following steps: S91. The dynamic parameters are called to update the data packet, and the speech recognition model is incrementally trained and optimized through machine learning algorithms. S92. Place the optimized model in a virtual BCM control scenario for simulation verification and generate strategy effectiveness verification results; S93. Integrate the verification results with the current model data, update and output the adaptive cooperative control model data.

Images