A method and related equipment for identifying hot words in vehicle scenarios

Abstract

This application discloses a method and related equipment for identifying hot words in a vehicle-mounted scene, relating to the field of Internet technology. The method includes: acquiring user-input voice data; extracting hot word text from the voice data; and acquiring fusion features between the voice data and the hot word text; parsing the hot word text to generate multiple mapping paths; constructing a hot word language model based on the multiple mapping paths; decoding the fusion features based on the hot word language model to obtain a hot word decoding result; decoding the fusion features based on the general language model to obtain a general decoding result; and determining a target recognition result based on the hot word decoding result and the general decoding result.

Description

Technical Field

[0001] This application relates to the field of Internet technology, and in particular to a method and related equipment for identifying hot words in vehicle scenarios. Background Technology

[0002] Currently, with the development of speech recognition technology, the recognition effect in general fields has been significantly improved. Under ideal conditions, the recognition accuracy can typically reach 95% or even higher. Meanwhile, speech recognition technology has been widely applied in the automotive field to realize functions such as in-vehicle infotainment systems, navigation, and voice assistants.

[0003] However, hot word recognition, such as names of people and places, and named entities, which have low word frequencies or are strongly relevant to users, cannot achieve the same accuracy as general recognition methods. Therefore, it is necessary to propose a hot word recognition method for in-vehicle scenarios to at least solve some of the above-mentioned problems. Summary of the Invention

[0004] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0005] In a first aspect, embodiments of this application provide a method for identifying hot words in a vehicle-mounted scene, the method comprising:

[0006] Acquire user-input voice data, extract hot word text from the voice data, and obtain the fusion features of the voice data and the hot word text;

[0007] The hot word text is parsed to generate multiple mapping paths, and a hot word language model is constructed based on the multiple mapping paths;

[0008] The fusion features are decoded based on the hot word language model to obtain the hot word decoding result;

[0009] The fusion features are decoded based on the general language model to obtain a general decoding result;

[0010] The target recognition result is determined based on the hot word decoding result and the general decoding result.

[0011] In one embodiment of the present invention, the step of obtaining the fusion features of the speech data and the hot word text includes:

[0012] The speech data is encoded to obtain a speech embedding vector;

[0013] The hot word text is encoded to obtain a text embedding vector;

[0014] The relevance of the speech embedding vector and the text embedding vector is calculated to obtain the relevance weight;

[0015] The speech embedding vector and the text embedding vector are fused based on the relevance weights to obtain fused features.

[0016] In one embodiment of the present invention, the step of fusing the speech embedding vector and the text embedding vector based on the relevance weight to obtain the fused feature includes:

[0017] In the absence of the hot word text, the speech embedding vector is used as the fusion feature.

[0018] In one embodiment of the present invention, the step of constructing a hot word language model based on the plurality of mapping paths includes:

[0019] The hot word text is parsed to obtain multiple hot word mapping relationships;

[0020] Multiple hot word mapping paths are constructed based on the aforementioned hot word mapping relationships;

[0021] By connecting multiple hot word mapping paths in parallel, an initial hot word language model is obtained;

[0022] Obtain the sentence text associated with the hot word text;

[0023] The sentence text is parsed to obtain multiple sentence mapping relationships;

[0024] Multiple sentence mapping paths are constructed based on the multiple sentence mapping relationships described above;

[0025] By connecting multiple sentence pattern mapping paths in parallel, a sentence pattern language model is obtained;

[0026] The sentence-based language model and the initial hot word language model are connected end to end to form a hot word language model.

[0027] In one embodiment of the present invention, the step of determining the target recognition result based on the hot word decoding result and the general decoding result includes:

[0028] The hot word path score and hot word pronunciation sequence distance are calculated from the hot word decoding results. The hot word path score includes the hot word acoustic score and the hot word language score.

[0029] The general path score and the general pronunciation sequence distance are calculated from the general decoding results. The general path score includes a general acoustic score and a general language score.

[0030] If the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and the difference between the language score of the hot word and the general language score is greater than a second preset threshold, the similarity between the pronunciation sequence distance of the hot word and the pronunciation sequence distance is determined.

[0031] If the similarity is greater than or equal to the third preset threshold, the hot word decoding result is used as the recognition result.

[0032] In one embodiment of the present invention, the step following the determination of the similarity between the hot word pronunciation sequence distance and the general pronunciation sequence distance includes:

[0033] If the similarity is less than a third preset threshold, determine whether the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and whether the difference between the language score of the hot word and the language score of the general language is greater than a second preset threshold.

[0034] If the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and the difference between the language score of the hot word and the general language score is greater than a second preset threshold, the hot word decoding result is used as the recognition result.

[0035] If the difference between the acoustic score of the hot word and the general acoustic score is less than a first preset threshold, or if the difference between the language score of the hot word and the general language score is less than a second preset threshold, the general decoding result shall be used as the recognition result.

[0036] In one embodiment of the present invention, the step of determining the target recognition result based on the hot word decoding result and the general decoding result further includes:

[0037] In the absence of the aforementioned hot word text, the general decoding result is used as the recognition result.

[0038] Secondly, this application proposes a hot word recognition system for in-vehicle scenarios, the system comprising: a data acquisition module, a parsing module, and a recognition module;

[0039] The data acquisition module is configured to: acquire user-input voice data, extract hot word text from the voice data, and acquire the fusion features of the voice data and the hot word text;

[0040] The parsing module is configured to: parse the hot word text, generate multiple mapping paths, and construct a hot word language model based on the multiple mapping paths;

[0041] The recognition module is configured to: decode the fusion features based on the hot word language model to obtain a hot word decoding result; decode the fusion features based on the general language model to obtain a general decoding result; and determine the target recognition result based on the hot word decoding result and the general decoding result.

[0042] Thirdly, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of a method for recognizing hot words in a vehicle scene as described in any of the first aspects above.

[0043] Fourthly, this application also proposes a computer-readable storage medium storing a computer program thereon, wherein when the computer program is executed by a processor, it implements the steps of the method for recognizing hot words in a vehicle scene according to any one of the first aspects.

[0044] In summary, the method for identifying hot words in a vehicle scene according to the embodiments of this application generates fixed contextual dependencies by constructing a hot word language model based on the mapping path, thereby improving the overall detection of hot words. Furthermore, it determines the target recognition result based on the hot word decoding result and the general decoding result, thereby reducing interference with the general recognition and improving the accuracy of hot word recognition.

[0045] The method for identifying hot words in vehicle scenarios proposed in this application, along with other advantages, objectives, and features of this application, will be partly apparent from the following description and partly understood by those skilled in the art through research and practice of this application. Attached Figure Description

[0046] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0047] Figure 1 A flowchart illustrating a method for identifying hot words in a vehicle-mounted scene, provided in an embodiment of this application;

[0048] Figure 2 This application provides a schematic diagram of the structure of a vehicle-mounted scene hot word recognition system.

[0049] Figure 3 This is a schematic diagram of the structure of an electronic device for recognizing and controlling hot words in a vehicle scene, provided in an embodiment of this application. Detailed Implementation

[0050] To better understand the technical solutions provided in the embodiments of this specification, the technical solutions of the embodiments of this specification will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this specification and the specific features in the embodiments are detailed descriptions of the technical solutions of the embodiments of this specification, rather than limitations on the technical solutions of this specification. In the absence of conflict, the embodiments of this specification and the technical features in the embodiments can be combined with each other.

[0051] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying 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 limitation, 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. The term "two or more" includes two or more cases.

[0052] Please see Figure 1 This is a flowchart illustrating a method for identifying hot words in a vehicle-mounted scene, provided in an embodiment of this application. Specifically, it may include:

[0053] S110. Obtain voice data input by the user, extract hot word text from the voice data, and obtain the fusion feature of the voice data and the hot word text.

[0054] For example, when a user speaks, the device's microphone collects the user's voice data. This voice data is then analyzed and processed, specifically: speech recognition technology is used to convert the speech into text; then, natural language processing algorithms (such as keyword extraction algorithms) are used to identify hot words from the converted text, and these extracted hot words are combined to form hot word text. Finally, the fusion features of the voice data and the hot word text are obtained.

[0055] S120. The hot word text is parsed to generate multiple mapping paths, and a hot word language model is constructed based on the multiple mapping paths.

[0056] For example, after obtaining the hot word text, it needs to be parsed to generate multiple mapping paths. These mapping paths can be understood as connections from the hot word text to different concepts or entities. For the hot word text "theme park," possible mapping paths include: pointing to the name of a specific well-known theme park (e.g., Disney theme park), pointing to the type of theme park (e.g., water park, adventure park), and pointing to activities related to the theme park (e.g., roller coasters, carousels). These mapping paths help the system to more comprehensively understand the meaning of the hot word text. A hot word language model is then constructed using these generated mapping paths.

[0057] S130. Decode the fusion features based on the hot word language model to obtain the hot word decoding result.

[0058] For example, the fused feature is decoded using a hot word language model to obtain the hot word decoding result. The decoding process involves using the hot word language model to analyze and understand the fused feature. For instance, if the hot word text corresponding to the fused feature is "park", the hot word decoding result might be a specific park name (such as "Central Park"), a description of the park type (such as "large urban park"), or a recommendation of park-related activities (such as "parks suitable for walking").

[0059] S140. Decode the fusion features based on the general language model to obtain a general decoding result.

[0060] For example, a generalized language model is used to decode the fused feature, yielding a generalized decoding result. When decoding a fused feature using a generalized language model, the model attempts to analyze and interpret the feature based on its learned language knowledge and patterns. The generalized language model takes a broader linguistic perspective, not just limited to specific hot words. For instance, if the speech data corresponding to the fused feature is "I want to go to that place to play," the generalized decoding result would be "The user expressed a desire to visit a specific location."

[0061] S150. Determine the target recognition result based on the hot word decoding result and the general decoding result.

[0062] For example, the target recognition result is determined based on the hot word decoding result and the general decoding result. The target recognition result is the final judgment obtained by comprehensively considering the hot word decoding result and the general decoding result, so as to obtain a more accurate and targeted result.

[0063] In summary, the hot word recognition method for in-vehicle scenarios proposed in this application improves the overall detection of hot words by generating fixed contextual dependencies through a hot word language model built based on the mapping path, and determines the target recognition result based on the hot word decoding result and the general decoding result, thereby reducing interference with the general recognition and improving the accuracy of hot word recognition.

[0064] In some examples, the step of obtaining the fusion features of the speech data and the hot word text includes:

[0065] The speech data is encoded to obtain a speech embedding vector;

[0066] The hot word text is encoded to obtain a text embedding vector;

[0067] The relevance of the speech embedding vector and the text embedding vector is calculated to obtain the relevance weight;

[0068] The speech embedding vector and the text embedding vector are fused based on the relevance weights to obtain fused features.

[0069] For example, speech data and hot word text are encoded separately using text injection to obtain speech embedding vectors and text embedding vectors. Text injection is a technique that introduces textual information during the encoding of speech data. Specifically, when encoding speech data and hot word text, speech-related text (such as transcribed text of speech content, related hot word text, etc.) is input as additional input or features, along with the original speech data and hot word text, into the encoding model.

[0070] The encoding process typically employs deep learning models, such as convolutional neural networks, recurrent neural networks, or Transformer architectures. These models learn the correspondence between large amounts of speech data and related text, transforming the speech data into a fixed-length speech embedding vector. This vector captures various features of the speech, including semantics, timbre, and intonation, while also incorporating additional information from text injection. Simultaneously, the hot-word text needs to be converted into a text embedding vector that represents its semantic and syntactic features.

[0071] Calculating the relevance between speech embedding vectors and text embedding vectors is to determine the degree of association between speech data and the hot word text. Various methods can be used to calculate relevance, such as calculating the cosine similarity between vectors or the Pearson correlation coefficient. These methods measure the directional similarity or the strength of the linear relationship between two vectors. The resulting relevance weight is a numerical value that reflects the closeness between the speech data and the hot word text. A high relevance results in a larger weight value, and vice versa.

[0072] The speech embedding vector and text embedding vector are fused based on relevance weights to obtain a fused feature. The fusion process aims to integrate information from speech data and hot word text to obtain a more comprehensive feature representation. The relevance weights determine the relative importance of the speech and text embedding vectors during the fusion process. A larger relevance weight indicates a high correlation between the speech and hot word text, allowing for greater weighting of the text embedding vector during fusion, making the fused feature more text-oriented. Conversely, a smaller relevance weight allows for greater weighting of the speech embedding vector. The fusion method can be a simple weighted sum or a more complex neural network structure. The resulting fused feature is equivalent to a hot word-related bias vector, enhancing the acoustic bias information of the hot words.

[0073] In some examples, the step of fusing the speech embedding vector and the text embedding vector based on the relevance weights to obtain fused features includes:

[0074] In the absence of the hot word text, the speech embedding vector is used as the fusion feature.

[0075] For example, if there is no hot word text and the relevance weight is 0, then the speech embedding vector obtained by encoding the speech data is directly used as the fusion feature. Normally, the speech data and hot word text are encoded separately, then the relevance weight is calculated and fused to obtain the fusion feature. However, when no hot word text is available, the fusion operation cannot be performed according to the normal process. In this case, to ensure that the system still has some usable feature representation, the only available speech embedding vector is directly used as the fusion feature. This allows subsequent processing tasks to be performed using the encoded speech data even without hot word text, avoiding the situation where effective features cannot be generated due to the lack of hot word text.

[0076] In some examples, the step of constructing a hot word language model based on multiple said mapping paths includes:

[0077] The hot word text is parsed to obtain multiple hot word mapping relationships;

[0078] Multiple hot word mapping paths are constructed based on the aforementioned hot word mapping relationships;

[0079] By connecting multiple hot word mapping paths in parallel, an initial hot word language model is obtained;

[0080] Obtain the sentence text associated with the hot word text;

[0081] The sentence text is parsed to obtain multiple sentence mapping relationships;

[0082] Multiple sentence mapping paths are constructed based on the multiple sentence mapping relationships described above;

[0083] By connecting multiple sentence pattern mapping paths in parallel, a sentence pattern language model is obtained;

[0084] The sentence-based language model and the initial hot word language model are connected end to end to form a hot word language model.

[0085] For example, in speech recognition based on a hybrid encoder-decoder structure, the language model can employ a decoder, a Finite State Transducer (FST), or similar methods. Considering the deterministic nature of the FST decoding path, this scheme uses FST as the language model. For general-domain FSTs, probabilistic statistics based on general corpora are used; however, for hot words in the automotive domain, effective path mapping needs to be manually constructed to generate hot word FSTs.

[0086] The hot word text is parsed to obtain multiple hot word mapping relationships. This parsing process involves semantic analysis and part-of-speech tagging for each hot word to determine the relationships between them. For example, some hot words may have semantic similarities, antonyms, or inclusion relationships. Ultimately, multiple hot word mapping relationships are obtained, which can be represented in the form of "hot word A and hot word B are semantically similar."

[0087] We can construct paths using the previously obtained hot word mapping relationships. For example, if hot word A and hot word B are semantically similar, we can construct a mapping path from hot word A to hot word B. Since there are multiple hot word mapping relationships, we can construct multiple such mapping paths.

[0088] By combining multiple hot word mapping paths into a network-like structure, with no specific order between the paths (similar to parallel connections in an electrical circuit), this network structure forms the initial hot word language model. It can be used to represent the relationships and possible combinations of hot words.

[0089] Simultaneously, the sentence structure texts related to the context of the hot word texts are collected. Similar to the parsing of the hot word texts, the sentence structure texts are analyzed to determine the relationships between different parts of the sentence structure, resulting in multiple sentence structure mapping relationships. These mapping relationships are then used to construct sentence structure mapping paths similar to the hot word mapping paths. Similar to constructing the initial hot word language model, multiple sentence structure mapping paths are connected in parallel to form a sentence structure language model. This model can represent the relationships and combinations between different sentence structure parts. Finally, the sentence structure language model and the initial hot word language model are concatenated to form a complete hot word language model with sentence structures.

[0090] In some examples, the step of determining the target recognition result based on the hot word decoding result and the general decoding result includes:

[0091] The hot word path score and hot word pronunciation sequence distance are calculated from the hot word decoding results. The hot word path score includes the hot word acoustic score and the hot word language score.

[0092] The general path score and the general pronunciation sequence distance are calculated from the general decoding results. The general path score includes a general acoustic score and a general language score.

[0093] If the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and the difference between the language score of the hot word and the general language score is greater than a second preset threshold, the similarity between the pronunciation sequence distance of the hot word and the pronunciation sequence distance is determined.

[0094] If the similarity is greater than or equal to the third preset threshold, the hot word decoding result is used as the recognition result.

[0095] For example, the hot word decoding results are used to calculate the hot word path score and the hot word pronunciation sequence distance. The hot word path score includes the hot word acoustic score and the hot word linguistic score. A higher hot word acoustic score indicates a better match between the hot word's acoustic features and the model. A higher hot word linguistic score indicates that the hot word is more in line with expectations at the linguistic level. A smaller hot word pronunciation sequence distance indicates that the hot word's pronunciation is closer to standard pronunciation. The general decoding results are also used to calculate the general path score and the general pronunciation sequence distance. The general path score includes the general acoustic score and the general linguistic score.

[0096] First, it is determined whether the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and whether the difference between the language score of the hot word and the general language score is greater than a second preset threshold. If both conditions are met, it indicates that the hot word differs significantly from the general data in terms of acoustics and language. If the above conditions are met, the similarity between the pronunciation sequence distance of the hot word and the pronunciation sequence distance of the general pronunciation is further determined. If the similarity between the pronunciation sequence distance of the hot word and the pronunciation sequence distance of the general pronunciation is greater than or equal to a third preset threshold, it indicates that although the pronunciation of the hot word differs from the pronunciation of the general data, the difference is within an acceptable range. In this case, the hot word decoding result is output as the final recognition result, meaning that the hot word decoding result is considered more reliable.

[0097] In some examples, the step following the determination of the similarity between the hot word pronunciation sequence distance and the general pronunciation sequence distance includes:

[0098] If the similarity is less than a third preset threshold, determine whether the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and whether the difference between the language score of the hot word and the language score of the general language is greater than a second preset threshold.

[0099] If the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and the difference between the language score of the hot word and the general language score is greater than a second preset threshold, the hot word decoding result is used as the recognition result.

[0100] If the difference between the acoustic score of the hot word and the general acoustic score is less than a first preset threshold, or if the difference between the language score of the hot word and the general language score is less than a second preset threshold, the general decoding result shall be used as the recognition result.

[0101] For example, when the similarity between the pronunciation sequence distance of a hot word and the general pronunciation sequence distance is less than a third preset threshold, it means that the pronunciation difference between the hot word and the general data exceeds an acceptable range. In this case, it is necessary to further determine the differences between the hot word and the general data in terms of acoustics and language. Specifically, this involves determining whether the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and whether the difference between the language score of the hot word and the general language score is greater than a second preset threshold.

[0102] If the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and the difference between the language score of the hot word and the general language score is greater than a second preset threshold, it indicates that the hot word differs significantly from the general data in terms of acoustics and language. In this case, the hot word decoding result is used as the recognition result. This is because hot words have special acoustic and linguistic features that are significantly different from the general data, making the hot word decoding result more reliable.

[0103] If the difference between the acoustic score of the hot word and the general acoustic score is less than a first preset threshold, or the difference between the language score of the hot word and the general language score is less than a second preset threshold, it indicates that the difference between the hot word and the general data in terms of acoustics or language is not significant enough. In this case, the general decoding result is used as the recognition result. This is because the difference between the hot word and the general data is not obvious, and the general decoding result is more universal and reliable.

[0104] In some examples, the step of determining the target recognition result based on the hot word decoding result and the general decoding result further includes:

[0105] In the absence of the aforementioned hot word text, the general decoding result is used as the recognition result.

[0106] For example, if no hot word text is currently available, the general decoding result is directly output as the final recognition result. If no hot word text appears, no special processing is needed for hot words; the more general decoding result can be used directly, since there is no specific hot word text to prioritize as the recognition result.

[0107] like Figure 2 As shown, this application proposes a system for recognizing hot words in vehicle scenarios. The system includes: a data acquisition module 21, a parsing module 22, and a recognition module 23.

[0108] The data acquisition module 21 is configured to: acquire user-input voice data, extract hot word text from the voice data, and acquire the fusion features of the voice data and the hot word text;

[0109] The parsing module 22 is configured to: parse the hot word text, generate multiple mapping paths, and construct a hot word language model based on the multiple mapping paths;

[0110] The recognition module 23 is configured to: decode the fusion features based on the hot word language model to obtain a hot word decoding result; decode the fusion features based on the general language model to obtain a general decoding result; and determine the target recognition result based on the hot word decoding result and the general decoding result.

[0111] The effects of applying the aforementioned method in the above system can be found in the description of the aforementioned method embodiments, and will not be repeated here.

[0112] like Figure 3 As shown, this application embodiment also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of any of the above-mentioned methods for identifying hot words in vehicle scenarios.

[0113] Since the electronic device described in this embodiment is the device used to implement the vehicle scene hot word recognition device in the embodiment of this application, those skilled in the art can understand the specific implementation method and its various variations of the electronic device in this embodiment based on the method described in the embodiment of this application. Therefore, how the electronic device implements the method in the embodiment of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiment of this application is within the scope of protection of this application.

[0114] In practical implementation, when the computer program 311 is executed by the processor, it can achieve the following: Figure 1 Any of the corresponding implementation methods in the embodiments.

[0115] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0116] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-readable program code.

[0117] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0118] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0119] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0120] This application also provides a computer program product, which includes computer software instructions that, when executed on a processing device, cause the processing device to execute the LDPC decoding method of a solid-state drive controller.

[0121] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0122] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0123] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0124] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0125] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0126] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0127] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

[0128] Although preferred embodiments have been described in this specification, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this specification.

[0129] Obviously, those skilled in the art can make various modifications and variations to this specification without departing from its spirit and scope. Therefore, if such modifications and variations fall within the scope of the claims and their equivalents, this specification is also intended to include such modifications and variations.

Claims

1. A method for recognizing hot words in a vehicle-mounted scene, characterized in that, The method includes: Acquire user-input voice data, extract hot word text from the voice data, and obtain the fusion features of the voice data and the hot word text; The hot word text is parsed to generate multiple mapping paths, and a hot word language model is constructed based on the multiple mapping paths; The fusion features are decoded based on the hot word language model to obtain the hot word decoding result; The fused features are decoded based on a general language model to obtain a general decoding result; The target recognition result is determined based on the hot word decoding result and the general decoding result; The steps for constructing a hot word language model based on multiple mapping paths include: The hot word text is parsed to obtain multiple hot word mapping relationships; Multiple hot word mapping paths are constructed based on the aforementioned hot word mapping relationships; By connecting multiple hot word mapping paths in parallel, an initial hot word language model is obtained; Obtain the sentence text associated with the hot word text; The sentence text is parsed to obtain multiple sentence mapping relationships; Multiple sentence mapping paths are constructed based on the multiple sentence mapping relationships described above; By connecting multiple sentence pattern mapping paths in parallel, a sentence pattern language model is obtained; The sentence-based language model and the initial hot word language model are connected end to end to form a hot word language model.

2. The method for recognizing hot words in vehicle scenarios according to claim 1, characterized in that, The step of obtaining the fusion features of the speech data and the hot word text includes: The speech data is encoded to obtain a speech embedding vector; The hot word text is encoded to obtain a text embedding vector; The relevance of the speech embedding vector and the text embedding vector is calculated to obtain the relevance weight; The speech embedding vector and the text embedding vector are fused based on the relevance weights to obtain fused features.

3. The method for recognizing hot words in vehicle scenarios according to claim 2, characterized in that, The step of fusing the speech embedding vector and the text embedding vector based on the relevance weight to obtain the fused feature includes: In the absence of the hot word text, the speech embedding vector is used as the fusion feature.

4. The method for recognizing hot words in vehicle scenarios according to claim 1, characterized in that, The step of determining the target recognition result based on the hot word decoding result and the general decoding result includes: The hot word path score and hot word pronunciation sequence distance are calculated from the hot word decoding results. The hot word path score includes the hot word acoustic score and the hot word language score. The general path score and the general pronunciation sequence distance are calculated from the general decoding results. The general path score includes a general acoustic score and a general language score. If the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and the difference between the language score of the hot word and the general language score is greater than a second preset threshold, the similarity between the pronunciation sequence distance of the hot word and the pronunciation sequence distance is determined. If the similarity is greater than or equal to the third preset threshold, the hot word decoding result is used as the recognition result.

5. The method for recognizing hot words in vehicle scenarios according to claim 4, characterized in that, The steps following the determination of the similarity between the hot word pronunciation sequence distance and the general pronunciation sequence distance include: If the similarity is less than a third preset threshold, determine whether the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and whether the difference between the language score of the hot word and the language score of the general language is greater than a second preset threshold. If the difference between the acoustic score of the hot word and the general acoustic score is greater than a first preset threshold, and the difference between the language score of the hot word and the general language score is greater than a second preset threshold, the hot word decoding result is used as the recognition result. If the difference between the acoustic score of the hot word and the general acoustic score is less than a first preset threshold, or if the difference between the language score of the hot word and the general language score is less than a second preset threshold, the general decoding result shall be used as the recognition result.

6. The method for recognizing hot words in vehicle scenarios according to claim 1, characterized in that, The step of determining the target recognition result based on the hot word decoding result and the general decoding result further includes: In the absence of the aforementioned hot word text, the general decoding result is used as the recognition result.

7. A system for recognizing hot words in vehicle scenarios, characterized in that, The system includes: a data acquisition module, a parsing module, and a recognition module; The data acquisition module is configured to: acquire user-input voice data, extract hot word text from the voice data, and acquire the fusion features of the voice data and the hot word text; The parsing module is configured to: parse the hot word text, generate multiple mapping paths, and construct a hot word language model based on the multiple mapping paths. The step of constructing the hot word language model based on the multiple mapping paths includes: parsing the hot word text to obtain multiple hot word mapping relationships; constructing multiple hot word mapping paths based on the multiple hot word mapping relationships; connecting the multiple hot word mapping paths in parallel to obtain an initial hot word language model; obtaining the sentence text associated with the hot word text; parsing the sentence text to obtain multiple sentence mapping relationships; constructing multiple sentence mapping paths based on the multiple sentence mapping relationships; connecting the multiple sentence mapping paths in parallel to obtain a sentence language model; and connecting the sentence language model and the initial hot word language model end-to-end to form a hot word language model. The recognition module is configured to: decode the fusion features based on the hot word language model to obtain a hot word decoding result; decode the fusion features based on a general language model to obtain a general decoding result; and determine the target recognition result based on the hot word decoding result and the general decoding result.

8. An electronic device, comprising: The memory and processor are characterized in that the processor is used to execute a computer program stored in the memory to implement the steps of a method for recognizing hot words in a vehicle scene as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for recognizing hot words in a vehicle scene as described in any one of claims 1-6.

Images