Multi-speech recognition method for intelligent automobile cabin and related equipment

Abstract

The invention provides a multi-speech recognition method for an intelligent automobile cabin and related equipment, and the method comprises the steps: decomposing the mixed frequencies of a plurality of sound signals from a sound mixed signal, and determining an initial speech recognition signal according to all the mixed frequencies; determining the noise tolerance in the automobile cabin through all the mixed frequencies and the vibration noise of the automobile in the driving process, and determining an effective voice section of the target user in the automobile cabin based on the noise tolerance and the sound mixed signal; extracting an effective recognition signal of a command sent by the target user from the initial voice recognition signal according to the effective voice section; and detecting a keyword of a trigger operation from the effective identification signal, and executing an operation instruction of a target user according to the keyword. According to the scheme, intelligent multi-voice recognition can be carried out on all received voice signals in the automobile cabin, the influence of noise of multiple ways on voice recognition in the driving process of an automobile is reduced, and the recognition accuracy of user commands in the automobile cabin is improved.

Description

Technical Field

[0001] The present application relates to the field of speech recognition technology, and more specifically, to a multi-speech recognition method for a smart car cockpit. Background Art

[0002] Speech recognition refers to the technology of computer software or computer systems automatically identifying and understanding human speech. It allows computers to convert speech into editable and storable text format by analyzing audio input. Speech recognition technology involves multiple complex steps such as sound signal processing, feature extraction, pattern matching and probability modeling.

[0003] Multi-speech recognition in the cockpit of an intelligent car is the ability to recognize, understand and process voice commands from multiple passengers in the car when there are multiple speakers in the cockpit. Compared with speech recognition of a single speaker, multi-speech recognition needs to distinguish and understand different speakers in a complex environment. In particular, when the car is moving, there are multiple sources of noise in the interior environment of the car cockpit, such as: noise from vehicle vibration, wind noise, voices of fellow passengers, etc., which causes the command signal received by the car to contain multiple sound signals, resulting in the car recognizing redundant command signals when performing speech recognition in response to user commands. Therefore, how to realize intelligent multi-speech recognition of all voice signals received in the car cockpit, reduce the impact of noise from multiple sources on speech recognition during driving, and improve the recognition accuracy of user commands in the car cockpit has become a difficult problem faced by the industry. Summary of the invention

[0004] The present application provides a multi-voice recognition method and related equipment for an intelligent car cockpit, which can realize intelligent multi-voice recognition of all voice signals received in the car cockpit, reduce the impact of noise from multiple channels on voice recognition during the driving process of the car, and improve the recognition accuracy of user commands in the car cockpit.

[0005] In a first aspect, the present application provides a multi-speech recognition method for a smart car cockpit, comprising the following steps: Collect all sound signals in the car cabin during driving to obtain a mixed sound signal in the car cabin; Decomposing a plurality of mixed frequencies of the sound signals from the sound mixed signal based on a preset decomposition scale, and determining an initial speech recognition signal of a target user in a vehicle cabin under a plurality of sound sources according to all the mixed frequencies; Determine the noise tolerance in the car cabin through all mixed frequencies and the vibration noise of the car during driving, perform speech enhancement recognition on the sound mixed signal based on the noise tolerance, and then obtain the effective speech segment of the target user in the car cabin; Extracting a valid recognition signal of a command issued by a target user from the initial voice recognition signal according to the valid voice segment; A keyword triggering an operation is detected from the effective recognition signal, and an operation instruction of a target user is executed according to the keyword.

[0006] In some embodiments, decomposing the mixed frequencies of the plurality of sound signals from the mixed sound signal based on a preset decomposition scale specifically includes: Determine the decomposition scale to be used in the speech recognition process; Decomposing the mixed sound signal according to the decomposition scale to obtain a plurality of high-frequency features and a plurality of low-frequency features; A mixed frequency of the plurality of sound signals is determined according to all high frequency characteristics and all low frequency characteristics.

[0007] In some embodiments, determining the initial speech recognition signal of the target user in the car cabin under multiple sound sources according to all mixed frequencies specifically includes: Determine the human voice recognition range in the car cabin; A plurality of high-frequency human voice distributions and a plurality of low-frequency human voice distributions in the vehicle cabin are determined according to the human voice recognition interval and all mixed frequencies.

[0008] The initial speech recognition signal of the target user under multiple sound sources is determined through the distribution of all high-frequency human voices and all low-frequency human voices.

[0009] In some embodiments, determining the noise tolerance in the vehicle cabin through all mixed frequencies and the vibration noise of the vehicle during driving specifically includes: Determine the signal fluctuation coefficient in the vehicle cabin based on all mixed frequencies; Determine the vibration and noise of the car during driving; The noise tolerance in the vehicle cabin is determined according to the signal fluctuation coefficient and the vibration noise.

[0010] In some embodiments, performing speech enhancement recognition on the sound mixed signal based on the noise tolerance to obtain the effective speech segment of the target user in the car cabin specifically includes: determining a plurality of enhancement signals according to the noise tolerance, all mixed frequencies and the sound mixed signal; Determine the priority coefficient of each reinforcement signal; The effective speech segment of the target user in the car cabin is extracted from all enhanced signals according to all priority coefficients.

[0011] In some embodiments, extracting the effective recognition signal of the target user's command from the initial voice recognition signal according to the effective voice segment specifically includes: Determining, according to the initial voice recognition signal, a plurality of initial signal segments when the target user issues a command; Determining the overlap of the valid speech segments through all the initial signal segments; A valid recognition signal of a command issued by a target user is determined according to the overlap degree and the valid voice segment.

[0012] In some embodiments, detecting the keyword that triggers the operation from the effective recognition signal is to convert the effective recognition signal into text, and use natural language processing technology to identify the keyword that triggers the operation in the text.

[0013] In a second aspect, the present application provides a multi-speech recognition system for a smart car cockpit, comprising: The acquisition module is used to collect all sound signals in the cabin of the car during driving to obtain a mixed sound signal in the cabin of the car; A processing module, configured to decompose the mixed frequencies of multiple sound signals from the mixed sound signal based on a preset decomposition scale, and determine the initial speech recognition signal of the target user in the vehicle cabin under multiple sound sources according to all the mixed frequencies; The processing module is further used to determine the noise tolerance in the car cabin through all mixed frequencies and the vibration noise of the car during driving, and perform voice enhancement recognition on the sound mixed signal based on the noise tolerance to obtain the effective voice segment of the target user in the car cabin; The processing module is further used to extract a valid recognition signal of a command issued by a target user from the initial voice recognition signal according to the valid voice segment; The execution module is used to detect a keyword that triggers an operation from the effective recognition signal, and execute an operation instruction of a target user according to the keyword.

[0014] In a third aspect, the present application provides a computer device, comprising a memory and a processor, wherein the memory stores codes, and the processor is configured to obtain the codes and execute the above-mentioned multi-voice recognition method for a smart car cockpit.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the multi-voice recognition method for the smart car cockpit is implemented.

[0016] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects: In the multi-speech recognition method and related equipment for an intelligent car cockpit provided by the present application, a mixed sound signal in the car cockpit is analyzed to determine a plurality of high-frequency frequency features and a plurality of low-frequency frequency features in the mixed sound signal, and then a plurality of mixed frequencies are determined, wherein the mixed frequencies are used to analyze the frequency features of the mixed sound signal, so as to extract the voice command of the target user in the car cockpit; secondly, all mixed frequencies are analyzed to determine the initial voice recognition signal of the target user in the car cockpit under a plurality of sound sources and the noise tolerance in the car cockpit, and then, the effective voice segment of the target user in the car cockpit is determined through the initial voice recognition signal and the noise tolerance, wherein the effective voice segment represents the voice signal with the highest signal strength emitted by the target user, and reduces the influence of noise from multiple paths during driving of the car on the voice recognition of the car; the effective recognition signal of the command emitted by the target user can be extracted from the initial voice recognition signal according to the effective voice segment, and finally, the keyword for triggering the operation is detected from the effective recognition signal, and the operation instruction of the target user is executed according to the keyword. The above scheme can realize intelligent multi-voice recognition of all voice signals received in the car cockpit, reduce the impact of noise from multiple channels on voice recognition during driving, and improve the recognition accuracy of user commands in the car cockpit. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is an exemplary flow chart of a multi-speech recognition method for a smart car cockpit according to some embodiments of the present application; Figure 2 is an exemplary flow chart of determining a mixing frequency according to some embodiments of the present application; Figure 3 is an exemplary flow chart of determining noise tolerance according to some embodiments of the present application; Figure 4 is a schematic diagram of the structure of a multi-speech recognition system for a smart car cockpit according to some embodiments of the present application; Figure 5 It is a structural schematic diagram of a computer device for implementing a multi-voice recognition method for a smart car cockpit according to some embodiments of the present application. DETAILED DESCRIPTION

[0018] In order to better understand the technical solution of the present application, the technical solution of the present application will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0019] refer to Figure 1 , which is an exemplary flow chart of a multi-speech recognition method for a smart car cockpit according to some embodiments of the present application. The multi-speech recognition method 100 for a smart car cockpit mainly includes the following steps: In step 101, all sound signals in the cabin of the car during driving are collected to obtain a mixed sound signal in the cabin of the car.

[0020] In a specific implementation, all sound signals in the cabin of the car during driving are collected by a sound collection device in the prior art, and the signal after superimposing all the sound signals is used as a mixed sound signal in the car cabin, wherein the sound collection device can be a vehicle-mounted microphone, a lavalier microphone, an acoustic analyzer, etc. It should be noted that the mixed sound signal is a superimposed signal composed of all sounds in the cabin of the car during driving, for example: the mixed sound signal includes signals such as engine noise, wind noise, and human voice.

[0021] In step 102, multiple mixed frequencies of sound signals are decomposed from the sound mixed signal based on a preset decomposition scale, and initial speech recognition signals of a target user in a car cabin under multiple sound sources are determined according to all the mixed frequencies.

[0022] In some embodiments, reference Figure 2 As shown in FIG. 1 , this figure is a schematic diagram of a flow chart of determining a mixed frequency in some embodiments of the present application. In this embodiment, decomposing the mixed frequencies of multiple sound signals from the sound mixed signal based on a preset decomposition scale can be implemented by the following steps: First, in step 1021, a decomposition scale preset in the speech recognition process is determined; Next, in step 1022, the sound mixed signal is decomposed according to the decomposition scale to obtain a plurality of high-frequency features and a plurality of low-frequency features; Finally, in step 1023, the mixed frequency of the plurality of sound signals is determined according to all the high-frequency characteristics and all the low-frequency characteristics.

[0023] In specific implementation, determining the decomposition scale preset in the speech recognition process can be achieved in the following manner, namely: extracting historical decomposition scale data from the database of the car, and using the historical decomposition scale with the highest frequency of use among all historical decomposition scales as the decomposition scale preset in the speech recognition process, wherein the decomposition scale represents the number of decomposition layers of the sound mixture signal, and is used to determine the number of decomposition times of the sound mixture signal; decomposing the sound mixture signal according to the decomposition scale to obtain a plurality of high-frequency frequency features and a plurality of low-frequency frequency features can be achieved in the following manner, namely: decomposing the sound mixture signal through a filter in the prior art in combination with a decomposition scale to obtain a plurality of low-frequency frequency features and a plurality of high-frequency frequency features in the sound mixture signal. For example, when the decomposition scale is 3, the number of decompositions is 3 times, and the first time, through The filter in the prior art decomposes the first low-frequency frequency feature and the first high-frequency frequency feature from the sound mixed signal. The second time, the filter in the prior art decomposes the second high-frequency frequency feature and the second low-frequency frequency feature from the first high-frequency frequency feature. The third time, the filter in the prior art decomposes the third high-frequency frequency feature and the third low-frequency frequency feature from the second high-frequency frequency feature. The high-frequency frequency feature represents the set of all frequency values ​​corresponding to the high-frequency part of the sound mixed signal, which is used to analyze the data details of the sound mixed signal. The low-frequency frequency feature represents the set of all frequency values ​​corresponding to the low-frequency part of the sound mixed signal, which is used to analyze the overall trend of the sound mixed signal. In other embodiments, other methods can also be used for determination, which are not limited here.

[0024] In specific implementation, the mixed frequencies of multiple sound signals can be determined based on all high-frequency frequency features and all low-frequency frequency features in the following manner, namely: selecting a high-frequency frequency feature as the selected high-frequency frequency feature, using the low-frequency frequency feature corresponding to the selected high-frequency frequency feature as the selected low-frequency frequency feature, using the set of all frequency values ​​corresponding to the selected high-frequency frequency feature and the selected low-frequency frequency feature as the mixed frequency of a sound signal, and continuing to determine the mixed frequencies of other sound signals. In other embodiments, other methods may also be used for determination, which are not limited here.

[0025] It should be noted that the mixed frequency in the present application represents the set of all frequency values ​​corresponding to the high-frequency frequency characteristics and the low-frequency frequency characteristics, which is used to analyze the frequency characteristics of the sound mixed signal to facilitate the extraction of the voice commands of the target user in the car cabin.

[0026] In some embodiments, determining the initial speech recognition signal of the target user in the car cabin under multiple sound sources based on all mixed frequencies can be implemented by the following steps: Determine the human voice recognition range in the car cabin; Determine multiple high-frequency human voice distributions and multiple low-frequency human voice distributions in the vehicle cabin according to the human voice recognition interval and all mixed frequencies; The initial speech recognition signal of the target user under multiple sound sources is determined through the distribution of all high-frequency human voices and all low-frequency human voices.

[0027] In specific implementation, the human voice recognition interval in the car cabin can be determined in the following manner, namely: extracting the human voice recognition interval in the car cabin from the car database, wherein the human voice recognition interval represents the frequency value interval when performing speech recognition on the human voice in the car cabin, and is used to filter the noise in the mixed sound signal; determining multiple high-frequency human voice distributions and multiple low-frequency human voice distributions in the car cabin according to the human voice recognition interval and all mixed frequencies can be implemented in the following manner, namely: selecting a mixed frequency as the selected mixed frequency, extracting all frequencies less than the upper limit of the human voice recognition interval from the high-frequency frequency characteristics of the selected mixed frequency, distributing all the extracted frequencies according to the corresponding time, and using the obtained frequency distribution as a high-frequency distribution of the sound emitted by the target user in the car cabin, and continuing to determine the car cabin A plurality of high-frequency distributions of the sound emitted by the target user in the cabin, wherein the high-frequency distribution represents the frequency distribution of the sound emitted by the target user after high-frequency noise is removed, and is used to identify the frequency of the human voice in the car cabin; a mixed frequency is selected as the selected mixed frequency, and all frequencies greater than the lower limit of the human voice recognition interval are extracted from the low-frequency frequency features in the selected mixed frequency, and all the extracted frequencies are distributed according to the corresponding acquisition time, and the obtained frequency distribution is used as a low-frequency distribution of the sound emitted by the target user in the car cabin, and a plurality of low-frequency distributions of the sound emitted by the target user in the car cabin are further determined, wherein the low-frequency distribution represents the frequency distribution of the sound emitted by the target user after low-frequency noise is removed, and is used to identify the frequency of the human voice in the car cabin; in other embodiments, other methods may also be used for determination, which are not limited here.

[0028] In specific implementation, determining the initial speech recognition signal of the target user under multiple sound sources through all high-frequency human voice distributions and all low-frequency human voice distributions can be achieved in the following manner, namely: selecting a high-frequency human voice distribution as the selected high-frequency human voice distribution, taking the selected high-frequency human voice distribution and the set of low-frequency human voice distributions corresponding to the selected high-frequency human voice distribution as the global spectrum corresponding to the selected high-frequency human voice distribution, and continuing to determine the global spectrum corresponding to the remaining high-frequency human voice distributions. The initial speech recognition signal of the target user under multiple sound sources can be determined by combining all the global spectrum with existing signal processing equipment (such as a digital signal processor, a field programmable gate array, an image processing unit, etc.). In other embodiments, other methods can also be used for determination, which is not limited here.

[0029] It should be noted that the global spectrum described in the present application represents the spectrum of low-frequency and high-frequency sounds emitted by the target user, and is used to identify the human voice signal of the target user; the initial voice recognition signal represents the initial signal of the car recognizing the sound emitted by the target user under multiple sound sources, and is used for the car to recognize the voice commands of the target user, so as to facilitate the extraction of voice commands in the target car.

[0030] In step 103, the noise tolerance in the car cabin is determined by all mixed frequencies and the vibration noise of the car during driving, and speech enhancement recognition is performed on the sound mixed signal based on the noise tolerance to obtain the effective speech segment of the target user in the car cabin.

[0031] In some embodiments, reference Figure 3 As shown, this figure is a schematic diagram of the process of determining noise tolerance in some embodiments of the present application. In this embodiment, the noise tolerance in the car cabin is determined by all mixed frequencies and the vibration noise of the car during driving, which can be achieved by the following steps: First, in step 1031, the signal fluctuation coefficient in the vehicle cabin is determined according to all mixed frequencies; Secondly, in step 1032, the vibration noise of the automobile during driving is determined; Finally, in step 1033, the noise tolerance in the vehicle cabin is determined according to the signal fluctuation coefficient and the vibration noise.

[0032] In specific implementation, determining the signal fluctuation coefficient in the car cabin according to all mixed frequencies can be achieved in the following manner, namely: extracting all high-frequency frequency features from all mixed frequencies, selecting a high-frequency frequency feature as the selected high-frequency frequency feature, dividing the standard deviation of all frequency values ​​in the selected high-frequency frequency feature by the mean of all frequency values ​​in the selected high-frequency frequency feature, and using the value obtained by the division as the signal fluctuation coefficient of the selected high-frequency frequency feature, and continuing to determine the signal fluctuation coefficients of the remaining high-frequency frequency features, and using the mean of the signal fluctuation coefficients of all high-frequency frequency features as the signal fluctuation coefficient in the car cabin, wherein the signal fluctuation coefficient is a parameter representing the fluctuation degree of the mixed sound signal collected by the car, and is used to analyze the fluctuation of the mixed sound signal; determining the vibration noise of the car during driving can be achieved in the following manner, namely: collecting the vibration noise of the car during driving through a vibration collection device (such as a laser vibrometer, a capacitive sensor, a vibration sensor) The noise tolerance in the vehicle cabin is determined according to the signal fluctuation coefficient and the vibration noise, and the noise tolerance in the vehicle cabin is determined by the following method, namely, initializing a noise tolerance model, taking the vibration noise as the initialization parameter of the noise tolerance model, taking the signal fluctuation coefficient as the constraint parameter of the noise tolerance model, and obtaining the noise tolerance in the vehicle cabin through the noise tolerance model, wherein the noise tolerance model is an adjustment model of noise tolerance established by a machine learning algorithm (such as a regression algorithm, a neural network, etc.), and the adjustment model is, for example: noise tolerance=vibration noise*A+signal fluctuation coefficient*B, wherein A and B are weight coefficients, and A and B can be learned through a training process, such as through a training data set, and can also be determined in other ways in other embodiments, which are not limited here.

[0033] It should be noted that the noise tolerance in the present application represents a parameter of the maximum tolerance to noise when performing speech recognition in a car cabin, and can be used to analyze the noise intensity in the sound mixed signal to facilitate denoising operations on the sound mixed signal.

[0034] In some embodiments, performing speech enhancement recognition on the sound mixed signal based on the noise tolerance to obtain the effective speech segment of the target user in the car cabin can be achieved by the following steps: determining a plurality of enhancement signals according to the noise tolerance, all mixed frequencies and the sound mixed signal; Determine the priority coefficient of each reinforcement signal; The effective speech segment of the target user in the car cabin is extracted from all enhanced signals according to all priority coefficients.

[0035] In specific implementation, the following method can be used to determine multiple enhanced signals based on the noise tolerance, all mixed frequencies and the sound mixed signal, namely: select one mixed frequency from all mixed frequencies as the selected mixed frequency, compare each frequency in the selected mixed frequency with the noise tolerance, if there is a frequency greater than the noise tolerance in the selected mixed frequency, then the mixed frequency is discarded, if each frequency in the selected mixed frequency is less than the noise tolerance, then the mixed frequency is retained, and an enhanced signal of the sound mixed signal is generated by combining the retained mixed frequency and the sound mixed signal through an existing signal processing device (such as a digital signal processor, a field programmable gate array, an image processing unit, etc.), and other enhanced signals of the sound mixed signal are continuously determined, wherein the enhanced signal represents a discrete signal decomposed from the sound mixed signal after noise is removed, and is used to screen the signal strength of the sound emitted by the target user in each enhanced signal. In other embodiments, other methods can also be used for determination, which are not limited here.

[0036] In specific implementation, determining the priority coefficient of each enhanced signal can be achieved in the following manner, namely: selecting one enhanced signal from all enhanced signals as the selected enhanced signal, subtracting the mean of all amplitudes in the selected enhanced signal from each amplitude in the selected enhanced signal, adding the absolute values ​​of all subtracted values, dividing the added value by the standard deviation of all amplitudes in the selected enhanced signal, using the divided value as the priority coefficient of the selected enhanced signal, and continuing to determine the priority coefficients of the remaining enhanced signals; extracting the effective voice segment of the target user in the car cabin from all enhanced signals according to all priority coefficients can be achieved in the following manner, namely: extracting the maximum priority coefficient from all priority coefficients, and using the enhanced signal corresponding to the maximum priority coefficient in all enhanced signals as the effective voice segment of the target user in the car cabin; in other embodiments, other methods can also be used for determination, which are not limited here.

[0037] It should be noted that the effective voice segment in the present application represents the voice signal with the highest signal strength emitted by the target user, which can be used to analyze the sound emitted by the target user, so as to facilitate the car to extract the voice command of the target user.

[0038] In step 104, a valid recognition signal of a command issued by a target user is extracted from the initial speech recognition signal according to the valid speech segment.

[0039] In some embodiments, extracting the effective recognition signal of the target user's command from the initial voice recognition signal according to the effective voice segment can be achieved by using the following steps: Determining, according to the initial voice recognition signal, a plurality of initial signal segments when the target user issues a command; Determining the overlap of the valid speech segments through all the initial signal segments; A valid recognition signal of a command issued by a target user is determined according to the overlap degree and the valid voice segment.

[0040] In a specific implementation, the following method can be used to determine the multiple initial signal segments of the valid voice segment based on the initial voice recognition signal, namely: a clustering algorithm in the prior art (such as K-means++, DBSCAN, spectral clustering, etc.) is combined with the initial voice recognition signal to obtain multiple initial signal segments when the target user issues a command, wherein the initial signal segment represents the signal segment after the signal of the target user issuing the command is divided, and is used to denoise the voice signal when the target user issues the command; the following method can be used to determine the overlapping area of ​​the valid voice segment through all the initial signal segments, namely: one initial signal segment among all the initial signal segments is selected as the selected initial signal segment, and an amplitude in the selected initial signal segment is selected As the selected amplitude, the amplitude that is the same as the selected amplitude is extracted from all the amplitudes, and the total number of all the extracted amplitudes is used as the similarity of the selected amplitude, and the similarities of the remaining amplitudes in the selected initial signal segment are continued to be determined, wherein the similarity represents the parameter value of the similarity of the amplitudes in the initial signal segment, all the similarities are added, and the added value is used as the overlap of the selected initial signal segment, and the overlap of the remaining initial signal segments is continued to be determined, and the average of all the overlaps is used as the overlap of the effective voice segment, wherein the overlap represents the parameter value of the repetition degree of the amplitude in the effective voice segment, which can be used to perform noise reduction analysis on the signal of the target user issuing the command; in other embodiments, other methods can also be used for determination, which are not limited here.

[0041] In specific implementation, determining the effective recognition signal of the command issued by the target user based on the overlap and the effective voice segment can be achieved in the following manner, namely: initializing an effective recognition signal model, taking the effective voice segment as the initialization parameter of the effective recognition signal model, taking the overlap as the constraint parameter of the effective recognition signal model, and obtaining the effective recognition signal of the command issued by the target user through the effective recognition signal model, wherein the effective recognition signal model can use a machine learning algorithm (such as a regression algorithm, a neural network, etc.) to establish an adjustment model of the effective recognition signal, and the adjustment model is, for example: effective recognition signal = effective voice segment * C + overlap * D, wherein C and D are weight coefficients, and the weight coefficients C and D can be learned through a training process, for example, through a training data set. In other embodiments, other methods can also be used to determine, which is not limited here.

[0042] It should be noted that the effective recognition signal in the present application represents the voice signal after noise reduction of the voice command issued by the target user, which can be used to extract the command issued by the target user, so as to facilitate the car to recognize the voice command text of the target user.

[0043] In step 105, a keyword triggering an operation is detected from the effective recognition signal, and an operation instruction of the target user is executed according to the keyword.

[0044] It should be noted that the keyword for triggering the operation in the present application refers to the keyword for triggering the corresponding operation of the car. For example, the keyword for triggering the operation is "turn on the air conditioner", etc., which is not specifically limited here. In some embodiments, the keyword for triggering the operation is detected from the effective recognition signal, and the operation instruction of the target user is executed according to the keyword. Specifically, the following steps can be adopted, namely: Convert the effective recognition signal into text, use natural language processing (NLP) technology to identify the keywords in the text that trigger the operation. After the keywords that trigger the operation are identified, they need to be matched with predefined instruction sets. These instruction sets may be stored in a database and are associated with specific operations or functions. The matching process between the keywords that trigger the operation and the instruction set may be based on exact matching, fuzzy matching or pattern matching of keywords, which is not specifically limited here. After the keywords that trigger the operation match the predefined instruction set, the car can be controlled to perform the corresponding operation according to the predefined instruction set, thereby realizing the execution of the target user's operation instructions.

[0045] In addition, in another aspect of the present application, in some embodiments, the present application provides a multi-speech recognition system for a smart car cockpit, referring to Figure 4 , which is a schematic diagram of the structure of a multi-speech recognition system for a smart car cockpit according to some embodiments of the present application. The multi-speech recognition system 400 for a smart car cockpit includes: a collection module 401, a processing module 402 and an execution module 403, which are described as follows: The acquisition module 401 in the present application is mainly used to collect all sound signals in the cabin of the car during driving, and obtain a mixed sound signal in the cabin of the car; Processing module 402, in the present application, is used to decompose the mixed frequencies of multiple sound signals from the sound mixed signal based on a preset decomposition scale, and determine the initial speech recognition signal of the target user in the car cabin under multiple sound sources according to all the mixed frequencies; It should be noted that the processing module 402 in the present application is also used to determine the noise tolerance in the car cabin through all mixed frequencies and the vibration noise of the car during driving, and perform voice enhancement recognition on the sound mixed signal based on the noise tolerance, so as to obtain the effective voice segment of the target user in the car cabin; In addition, it should be noted that the processing module 402 in the present application is also used to extract the effective recognition signal of the target user's command from the initial voice recognition signal according to the effective voice segment; The execution module 403 in the present application is mainly used to detect the keyword that triggers the operation from the effective recognition signal, and execute the operation instruction of the target user according to the keyword.

[0046] In addition, the present application also provides a computer device, which includes a memory and a processor, the memory stores a code, and the processor is configured to obtain the code and execute the above-mentioned multi-voice recognition method for the smart car cockpit.

[0047] In some embodiments, reference Figure 5 , which is a schematic diagram of the structure of a computer device for implementing a multi-speech recognition method for a smart car cockpit according to some embodiments of the present application. The multi-speech recognition method for a smart car cockpit in the above embodiment can be Figure 5 The computer device 500 shown in the figure is implemented, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503 and at least one communication interface 504.

[0048] The processor 501 may be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0049] The communication bus 502 may be used to transmit information between the above-mentioned components.

[0050] The memory 503 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including a compressed optical disc, a laser disc, an optical disc, a digital versatile disc, a Blu-ray disc, etc.), a magnetic disk or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 503 may exist independently and be connected to the processor 501 via the communication bus 502. The memory 503 may also be integrated with the processor 501.

[0051] The memory 503 is used to store the program code for executing the solution of the present application, and the execution is controlled by the processor 501. The processor 501 is used to execute the program code stored in the memory 503. The program code may include one or more software modules. The method used in the above embodiment can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0052] The communication interface 504 uses any transceiver or other device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0053] In a specific implementation, as an embodiment, a computer device may include multiple processors, each of which may be a single-CPU processor or a multi-CPU processor. The processor here may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0054] The above-mentioned computer device may be a general-purpose computer device or a special-purpose computer device. In a specific implementation, the computer device may be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device or an embedded device. The embodiment of the present application does not limit the type of computer device.

[0055] In addition, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the multi-voice recognition method for the smart car cockpit is implemented.

[0056] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0057] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A multi-speech recognition method for an intelligent vehicle cockpit, characterized in that: The steps include: Collect all sound signals in the car cabin during driving to obtain a mixed sound signal in the car cabin; Decomposing a plurality of mixed frequencies of the sound signals from the sound mixed signal based on a preset decomposition scale, and determining an initial speech recognition signal of a target user in a vehicle cabin under a plurality of sound sources according to all the mixed frequencies; Determine the noise tolerance in the car cabin through all mixed frequencies and the vibration noise of the car during driving, perform speech enhancement recognition on the sound mixed signal based on the noise tolerance, and then obtain the effective speech segment of the target user in the car cabin; Extracting a valid recognition signal of a command issued by a target user from the initial voice recognition signal according to the valid voice segment; A keyword triggering an operation is detected from the effective recognition signal, and an operation instruction of a target user is executed according to the keyword.

2. The method according to claim 1, characterized in that Decomposing the mixed frequencies of the plurality of sound signals from the mixed sound signal based on a preset decomposition scale specifically includes: Determine the decomposition scale to be used in the speech recognition process; Decomposing the mixed sound signal according to the decomposition scale to obtain a plurality of high-frequency features and a plurality of low-frequency features; A mixed frequency of the plurality of sound signals is determined according to all high frequency characteristics and all low frequency characteristics.

3. The method according to claim 1, characterized in that Determining the initial speech recognition signal of the target user in the car cabin under multiple sound sources based on all mixed frequencies specifically includes: Determine the human voice recognition range in the car cabin; Determine multiple high-frequency human voice distributions and multiple low-frequency human voice distributions in the vehicle cabin according to the human voice recognition interval and all mixed frequencies; The initial speech recognition signal of the target user under multiple sound sources is determined through the distribution of all high-frequency human voices and all low-frequency human voices.

4. The method according to claim 1, characterized in that The noise tolerance in the car cabin is determined by all mixed frequencies and the vibration noise of the car during driving, including: Determine the signal fluctuation coefficient in the vehicle cabin based on all mixed frequencies; Determine the vibration and noise of the car during driving; The noise tolerance in the vehicle cabin is determined according to the signal fluctuation coefficient and the vibration noise.

5. The method according to claim 1, characterized in that The voice enhancement recognition is performed on the sound mixed signal based on the noise tolerance, thereby obtaining the effective voice segment of the target user in the car cabin, specifically including: determining a plurality of enhancement signals according to the noise tolerance, all mixed frequencies and the sound mixed signal; Determine the priority coefficient of each reinforcement signal; The effective speech segment of the target user in the car cabin is extracted from all enhanced signals according to all priority coefficients.

6. The method according to claim 1, characterized in that Extracting the effective recognition signal of the target user's command from the initial voice recognition signal according to the effective voice segment specifically includes: Determining, according to the initial voice recognition signal, a plurality of initial signal segments when the target user issues a command; Determining the overlap of the valid speech segments through all the initial signal segments; A valid recognition signal of a command issued by a target user is determined according to the overlap degree and the valid voice segment.

7. The method according to claim 1, characterized in that Detecting the keyword triggering the operation from the effective recognition signal is to convert the effective recognition signal into text, and use natural language processing technology to identify the keyword triggering the operation in the text.

8. A multi-speech recognition system for an intelligent car cockpit, characterized in that: include: The acquisition module is used to collect all sound signals in the cabin of the car during driving to obtain a mixed sound signal in the cabin of the car; A processing module, configured to decompose the mixed frequencies of multiple sound signals from the mixed sound signal based on a preset decomposition scale, and determine the initial speech recognition signal of the target user in the vehicle cabin under multiple sound sources according to all the mixed frequencies; The processing module is further used to determine the noise tolerance in the car cabin through all mixed frequencies and the vibration noise of the car during driving, and perform voice enhancement recognition on the sound mixed signal based on the noise tolerance to obtain the effective voice segment of the target user in the car cabin; The processing module is further used to extract a valid recognition signal of a command issued by a target user from the initial voice recognition signal according to the valid voice segment; The execution module is used to detect a keyword that triggers an operation from the effective recognition signal, and execute an operation instruction of a target user according to the keyword.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores codes, and the processor is configured to obtain the codes and execute the multi-voice recognition method for a smart car cockpit as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the multi-voice recognition method for an intelligent vehicle cockpit as described in any one of claims 1 to 7 is implemented.

Images