Method and system for extracting voice abstract in vehicle

Abstract

The invention relates to the technical field of voice extraction, and discloses an in-vehicle voice abstract extraction method and device, and the method comprises the steps: obtaining an in-vehicle voice text, dividing the voice text according to a vehicle travel, and obtaining a first text sentence matrix; performing subject term extraction on the first text sentence matrix, and dividing the first text sentence matrix to obtain a plurality of second text sentence matrixes; performing word segmentation on the second text sentence matrix to obtain a word vector text, and determining a TF-IDF value of each word vector; marking position features of the second text sentence matrix to obtain a sentence position matrix; carrying out weighting on the sentences, and determining the weight of each sentence position; and determining a comprehensive score of each word vector according to the TF-IDF values of the word vectors and the weight of each sentence position, and performing abstract extraction according to the comprehensive scores. Text abstract extraction can be carried out at the vehicle end, the analysis efficiency is greatly improved, the efficiency is improved, and the data transmission cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of speech extraction, and in particular to a method and system for extracting speech summaries in a vehicle. Background Art

[0002] In modern vehicle intelligent systems, a huge amount of voice and text data is generated every day. These data include the communication between passengers in the car, the command confirmation of the vehicle system, and various other sound information. If all these voice data are uploaded to the server for processing without screening, although it seems comprehensive, it will actually cause several problems. First, such a large amount of data upload will significantly reduce the efficiency of data analysis, because processing and analyzing large amounts of data requires more time and computing resources. Secondly, a large amount of data transmission will also cause network delays, which is not conducive to vehicle systems that require real-time response. In addition, storing these massive amounts of data requires a lot of space, which is undoubtedly a waste when storage resources are limited. Therefore, there is an urgent need for a method and system for extracting in-vehicle voice summaries to achieve summary extraction of text data locally on the vehicle side. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for extracting in-vehicle voice summaries, so as to solve the problem that if all voice data are uploaded, the analysis efficiency will be reduced, delay will be generated, and space will be wasted.

[0004] The present invention provides a method for extracting a summary of in-vehicle speech, the method comprising:

[0005] Acquire in-vehicle voice text, and divide the voice text according to the vehicle journey to obtain a plurality of first text sentence matrices;

[0006] Extracting subject words from each first text sentence matrix, and dividing the first text sentence matrix according to the extracted subject words to obtain a plurality of second text sentence matrices;

[0007] Segmenting the second text sentence matrix to obtain word vector text, and determining the TF-IDF value of each word vector according to the word vector text;

[0008] Marking position features of the second text sentence matrix to obtain a sentence position matrix;

[0009] Weighting the sentence position matrix and determining the weight of each sentence position;

[0010] The comprehensive score of each word vector is determined according to the TF-IDF value of each word vector and the weight of each sentence position, and summary extraction is performed according to the comprehensive score.

[0011] Preferably, the speech text is divided according to the vehicle journey to obtain a plurality of first text sentence matrices, including:

[0012] Acquire a plurality of vehicle navigation routes, and determine a starting point and a corresponding arrival point of each of the vehicle navigation routes;

[0013] The speech text collected during the time from the vehicle's starting point to the corresponding arrival point is set as the first text sentence matrix.

[0014] Preferably, the subject word extraction adopts one of the TextRank algorithm, LDA model, BERTopic model and supervised and unsupervised extraction algorithms.

[0015] Preferably, the first text sentence matrix is ​​divided according to the extracted subject words to obtain a plurality of second text sentence matrices, including:

[0016] The speech texts in the first text sentence matrix are classified according to the extracted subject words, the speech texts belonging to the same subject word are divided into the same category, and a second text sentence matrix is ​​generated.

[0017] Preferably, marking the position features of the second text sentence matrix to obtain a sentence position matrix includes:

[0018] Determining the acquisition time of the speech text in the second text sentence matrix,

[0019] Determine to sort the speech text once according to the acquisition time;

[0020] When the acquisition time is the same, determining the word order position relationship of the speech text in the second text sentence matrix;

[0021] The speech texts acquired at the same time are sorted again according to the word order position relationship to obtain a sentence position matrix.

[0022] Preferably, weighting the sentence position matrix and determining the weight of each sentence position includes:

[0023] The sentence position matrix is ​​weighted to obtain the sentence position matrix:

[0024] Normalize the sentence position matrix to get the matrix Y i ;

[0025] According to the matrix Y i Get the horizontal information entropy E i and longitudinal information entropy E j ;

[0026] According to the horizontal information entropy Ei and longitudinal information entropy E j Determine the weight of each sentence position.

[0027] Preferably, the matrix Y i Each element Y ij Calculated according to the following formula:

[0028]

[0029] Among them, x ij is the sentence position matrix Xi and Y ij The element corresponding to the position, min(x i ) is the minimum value, max(x i ) is the maximum value.

[0030] Preferably, according to the lateral information entropy E i and longitudinal information entropy E j Determine the weight of each sentence position, including:

[0031] According to the horizontal information entropy E i Determine the horizontal weight w of each sentence position i , calculate the lateral weight w according to the following formula i ;

[0032]

[0033] According to the longitudinal information entropy E j Determine the vertical weight w of each sentence position j , calculate the longitudinal weight w according to the following formula j ;

[0034]

[0035] Preferably, a comprehensive score of each word vector is determined according to the TF-IDF value of each word vector and the weight of each sentence position, and abstract extraction is performed according to the comprehensive score, including:

[0036] Determine the speech text to which each word vector belongs, and determine the weight of the corresponding sentence position according to the speech text to which it belongs;

[0037] The comprehensive score of each word vector is determined based on the TF-IDF value of the word vector and the weight of the corresponding sentence position. The comprehensive score is calculated according to the following formula:

[0038] C score =TF×IDF×w i × j ;

[0039] A summary is extracted from the word vector according to the comprehensive score.

[0040] The present invention also discloses a system for extracting in-vehicle voice abstracts, which is used to apply the above-mentioned method for extracting in-vehicle voice abstracts. The system comprises:

[0041] A voice text acquisition module, the acquisition module is used to acquire the voice text in the car;

[0042] A sentence matrix division module, the sentence matrix division module is used to divide the speech text according to the vehicle journey to obtain a plurality of first text sentence matrices; and extract a subject word from each first text sentence matrix, and divide the first text sentence matrix according to the extracted subject words to obtain a plurality of second text sentence matrices;

[0043] A TF-IDF value determination module, the TF-IDF value determination module is used to segment the second text sentence matrix to obtain word vector text, and determine the TF-IDF value of each word vector according to the word vector text;

[0044] A sentence weight determination module, the sentence weight determination module is used to mark the position features of the second text sentence matrix to obtain a sentence position matrix; and to weight the sentence position matrix and determine the weight of each sentence position;

[0045] The summary extraction module is used to determine the comprehensive score of each word vector according to the TF-IDF value of each word vector and the weight of each sentence position, and perform summary extraction according to the comprehensive score.

[0046] Compared with the prior art, the beneficial effect of the present invention is that the present invention can extract the summary of the in-car voice in real time, providing the driver with a fast and accurate information overview. By extracting and dividing the subject words of the voice text, the generated text sentence matrix is ​​more targeted and hierarchical, and the efficiency of information processing is improved. The word segmentation and TF-IDF value calculation of the voice text further enhance the recognition ability of key information and ensure the rapid locking of core words among many words. The design of sentence weight determination not only takes into account the position characteristics of the sentence in the text, but also enables sentences in different positions to have different weights through the weighting method, thereby improving the accuracy and reliability of the summary extraction, and the real weight of the words can be calculated to improve the accuracy. Finally, the summary is extracted according to the comprehensive score to ensure that the summary content can fully and accurately reflect the core information of the original voice text. The present invention can extract text summaries on the vehicle side, compress the space occupied by data upload to the cloud, greatly improve the analysis efficiency, improve efficiency, and reduce data transmission costs.

[0047] In addition, the in-vehicle speech summary extraction system of the present invention is also highly scalable and customizable. Users can configure and adjust various modules of the system according to actual needs to meet different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0049] Figure 1 The present invention is a flowchart of a method for extracting in-vehicle speech abstracts.

[0050] Figure 2 The present invention is a functional block diagram of a system for extracting in-vehicle speech abstracts. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] like Figure 1 As shown, the present invention provides a method for extracting a summary of in-vehicle speech, the method comprising:

[0053] S1, obtaining in-vehicle voice text, dividing the voice text according to the vehicle journey, and obtaining a plurality of first text sentence matrices.

[0054] S2, extracting keywords from each first text sentence matrix, and dividing the first text sentence matrix according to the extracted keywords to obtain a plurality of second text sentence matrices.

[0055] S3, segmenting the second text sentence matrix to obtain word vector text, and determining the TF-IDF value of each word vector based on the word vector text.

[0056] S4, marking position features of the second text sentence matrix to obtain a sentence position matrix.

[0057] S5, weighting the sentence position matrix and determining the weight of each sentence position.

[0058] S6, determining a comprehensive score for each word vector according to the TF-IDF value of each word vector and the weight of each sentence position, and performing summary extraction according to the comprehensive score.

[0059] In some embodiments of the present application, the voice text is divided according to the vehicle journey to obtain a plurality of first text sentence matrices, including: obtaining a plurality of vehicle navigation routes, determining the starting point and the corresponding arrival point of each of the vehicle navigation routes; and setting the voice text collected during the time from the starting point to the corresponding arrival point of the vehicle as the first text sentence matrix.

[0060] In this embodiment, the navigation route information in the vehicle is collected, which records the path and destination of the vehicle in detail. For each navigation route, its starting position and the final destination are accurately determined. On this basis, the voice text collected during the period from the vehicle starting from the starting position to the predetermined destination is defined as the first text sentence matrix. Such a definition not only ensures the integrity and accuracy of the text matrix, but also provides a reliable data basis for subsequent voice analysis and processing.

[0061] In some embodiments of the present application, the topic word extraction adopts one of the TextRank algorithm, LDA model, BERTopic model and supervised and unsupervised extraction algorithms.

[0062] In this embodiment, subject word extraction is a key step, which helps to identify the most important concepts and topics from a large amount of text. To achieve this goal, a variety of advanced algorithms and technologies can be used, including TextRank algorithm, LDA (Latent Dirichlet Allocation) model, BERTopic model, and one of supervised and unsupervised extraction algorithms. TextRank algorithm is a keyword extraction method based on word co-occurrence relationship. It constructs a directed acyclic graph (DAG) by calculating the similarity between words in the text, and then ranks the nodes in the graph using the PageRank algorithm to obtain the most important keywords. This method is simple, efficient, easy to implement, and particularly suitable for processing shorter texts. The LDA model is a topic model that can assign documents in a text collection to different topics, and can also extract keywords related to these topics from the documents. The LDA model models the topic through Bayes' theorem, and can discover hidden structures in the text collection, thereby helping users understand and analyze the text content. The BERTopic model is a keyword extraction model based on BERT (Bidirectional Encoder Representations from Transformers). BERT is a pre-trained language representation model that can learn deep semantic information in text. The BERTopic model uses the text representation generated by BERT to extract representative keywords from the text through clustering and topic modeling techniques. One of the supervised and unsupervised extraction algorithms is also used for topic word extraction. The supervised extraction algorithm is a method of learning to extract keywords from text based on known labels or classification information. The unsupervised extraction algorithm does not rely on pre-set labels and automatically identifies topics from text through clustering, dimensionality reduction and other techniques. These two methods have their own advantages and disadvantages, and can be selected based on actual needs and text characteristics.

[0063] In some embodiments of the present application, the first text sentence matrix is ​​divided according to the extracted keywords to obtain several second text sentence matrices, including: classifying the speech texts in the first text sentence matrix according to the extracted keywords, dividing the speech texts belonging to the same keyword into the same category, and generating a second text sentence matrix.

[0064] In this embodiment, first, the subject words are extracted from the given first text sentence matrix. After the subject words are extracted, the first text sentence matrix is ​​divided according to these subject words, and a number of second text sentence matrices are obtained. These second text sentence matrices are all divided based on the first text sentence matrix, but are more refined and classified. Specifically, the speech texts in the first text sentence matrix are classified. During the classification process, the speech texts belonging to the same subject word are divided into the same category. The advantage of doing this is that it allows us to understand the speech text content under each subject word more clearly. After the classification is completed, a new matrix, the second text sentence matrix, is obtained.

[0065] In some embodiments of the present application, position features are marked on the second text sentence matrix to obtain a sentence position matrix, including: determining the acquisition time of the speech text in the second text sentence matrix; determining a sorting of the speech text according to the acquisition time; when the acquisition times are the same, determining the word order position relationship of the speech text in the second text sentence matrix; and performing a second sorting of the speech text at the same acquisition time according to the word order position relationship to obtain a sentence position matrix.

[0066] In this embodiment, for the second text sentence matrix mentioned, the acquisition time of the voice text in the second text sentence matrix is ​​accurately determined. This step is the basis and the key, because it is directly related to the subsequent sorting and determination of the position relationship. Next, according to the acquisition time of the acquired voice texts, they are sorted. This sorting is based on time, which means that those voice texts with earlier acquisition time will be ranked in front, and those with later acquisition time will be ranked in the back. However, when the acquisition time is the same, the word order position relationship of the voice text needs to be considered. This is because at the same time point, there may be multiple voice texts generated, and the order between them needs to be determined according to the word order position relationship. Finally, according to these word order position relationships, the voice texts with the same acquisition time are sorted twice. Through such processing, a complete sentence position matrix can be obtained, which clearly shows the position relationship of each voice text in the matrix. Due to the characteristics of Chinese habits, for the same problem, the more front content is more representative, and in the same sentence, the more front part, the more representative the content; therefore, this method is used for sorting and processing, and finally a sentence position matrix that can accurately reflect the position relationship of the voice text is obtained.

[0067] In some embodiments of the present application, weighting the sentence position matrix and determining the weight of each sentence position include:

[0068] The sentence position matrix is ​​weighted to obtain the sentence position matrix:

[0069] Normalize the sentence position matrix to get the matrix Y i ; According to the matrix Y i Get the horizontal information entropy E i and longitudinal information entropy E j According to the horizontal information entropy E i and longitudinal information entropy E j Determine the weight of each sentence position.

[0070] In this embodiment, the sentence position matrix is ​​weighted by using the entropy weight method. After the weighting is completed, a new sentence position matrix is ​​obtained, which we mark as X i Next, in order to ensure the reasonable distribution of weights, the weighted sentence position matrix is ​​normalized. Normalization is to ensure that the sum of all values ​​in the matrix is ​​1, so that the weights of each position can be compared with each other. After this step, the normalized matrix Y is obtained. i After obtaining the normalized matrix Y i After that, the lateral information entropy E is further calculated. i and longitudinal information entropy E j The information entropy of a matrix can be obtained by calculating the entropy value of the probability distribution of the elements in the matrix. Finally, according to the calculated horizontal information entropy E i and longitudinal information entropy E j To finally determine the weight of each sentence position.

[0071] In some embodiments of the present application, the matrix Y i Each element Y ij Calculated according to the following formula:

[0072]

[0073] Among them, x ij is the sentence position matrix Xi and Y ij The element corresponding to the position, min(x i ) is the minimum value, max(x i ) is the maximum value.

[0074] In this embodiment, normalization is a commonly used method in the field of data processing and machine learning, and its core goal is to standardize the data so as to effectively compare and analyze different data. By normalizing the data, the distribution of the data can be scaled to a fixed range, usually scaling the data to between 0 and 1. The purpose of this is to eliminate the influence of the dimensions and numerical values ​​between different features, so that the model can focus more on the structure and characteristics of the data itself.

[0075] In some embodiments of the present application, according to the lateral information entropy E i and longitudinal information entropy E j Determine the weight of each sentence position, including: according to the horizontal information entropy E i Determine the horizontal weight w of each sentence position i , calculate the lateral weight w according to the following formula i ;

[0076]

[0077] According to the longitudinal information entropy E j Determine the vertical weight w of each sentence position j , calculate the longitudinal weight w according to the following formula j ;

[0078]

[0079] In some embodiments of the present application, a comprehensive score of each word vector is determined according to the TF-IDF value of each word vector and the weight of each sentence position, and summary extraction is performed according to the comprehensive score, including: determining the speech text to which each word vector belongs, and determining the weight of the corresponding sentence position according to the speech text to which it belongs; determining the comprehensive score of the word vector according to the TF-IDF value of each word vector and the weight of the corresponding sentence position, and the comprehensive score is calculated according to the following formula:

[0080] C score =TF×IDF×w i × j ;

[0081] A summary is extracted from the word vector according to the comprehensive score.

[0082] In this embodiment, the comprehensive score of word vectors is obtained by improving the traditional TFIDF method. Summary extraction is performed based on the comprehensive scores of these word vectors. Specifically, those word vectors with higher comprehensive scores will be selected because they are more likely to represent the core content of the text. In this way, the most critical information can be extracted from the original text to generate a concise and accurate summary. This process not only improves the efficiency of information processing, but also enables users to quickly understand the main content of the text, which is of great value in many application scenarios.

[0083] like Figure 2 As shown, the present invention also discloses a system for extracting in-vehicle voice summaries, which is used to apply the above-mentioned method for extracting in-vehicle voice summaries, and the system comprises:

[0084] A voice text acquisition module is used to acquire the voice text in the car.

[0085] The sentence matrix division module is used to divide the speech text according to the vehicle journey to obtain a plurality of first text sentence matrices; and to extract the subject words from each first text sentence matrix, and to divide the first text sentence matrix according to the extracted subject words to obtain a plurality of second text sentence matrices.

[0086] The TF-IDF value determination module is used to segment the second text sentence matrix to obtain word vector text, and determine the TF-IDF value of each word vector according to the word vector text.

[0087] The sentence weight determination module is used to mark the position features of the second text sentence matrix to obtain a sentence position matrix; and to weight the sentence position matrix and determine the weight of each sentence position.

[0088] The summary extraction module is used to determine the comprehensive score of each word vector according to the TF-IDF value of each word vector and the weight of each sentence position, and perform summary extraction according to the comprehensive score.

[0089] The present invention can extract the summary of the in-car voice in real time, and provide the driver with a fast and accurate information overview. By extracting and dividing the subject words of the voice text, the generated text sentence matrix is ​​more targeted and hierarchical, and the efficiency of information processing is improved. The voice text is segmented and the TF-IDF value is calculated, which further enhances the recognition ability of key information and ensures that the core vocabulary is quickly locked in a large number of words. The design of sentence weight determination not only takes into account the position characteristics of the sentence in the text, but also through the weighting method, the sentences in different positions have different weights, thereby improving the accuracy and reliability of the summary extraction, and the real weight of the words can be calculated to improve the accuracy. Finally, the summary is extracted according to the comprehensive score to ensure that the summary content can comprehensively and accurately reflect the core information of the original voice text.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

[0091] The system provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiment can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.

[0092] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

Claims

1. A method for extracting in-vehicle speech summaries, characterized in that: The method comprises: Acquire in-vehicle voice text, and divide the voice text according to the vehicle journey to obtain a plurality of first text sentence matrices; Extracting subject words from each first text sentence matrix, and dividing the first text sentence matrix according to the extracted subject words to obtain a plurality of second text sentence matrices; Segmenting the second text sentence matrix to obtain word vector text, and determining the TF-IDF value of each word vector according to the word vector text; Marking position features of the second text sentence matrix to obtain a sentence position matrix; Weighting the sentence position matrix and determining the weight of each sentence position; The comprehensive score of each word vector is determined according to the TF-IDF value of each word vector and the weight of each sentence position, and summary extraction is performed according to the comprehensive score.

2. The method for extracting in-vehicle speech abstracts according to claim 1, characterized in that: The speech text is divided according to the vehicle journey to obtain a plurality of first text sentence matrices, including: Acquire a plurality of vehicle navigation routes, and determine a starting point and a corresponding arrival point of each of the vehicle navigation routes; The speech text collected during the time from the vehicle's starting point to the corresponding arrival point is set as the first text sentence matrix.

3. The method for extracting in-vehicle speech abstracts according to claim 1, characterized in that: The topic word extraction adopts one of the TextRank algorithm, LDA model, BERTopic model and supervised and unsupervised extraction algorithms.

4. The method for extracting in-vehicle speech abstracts according to claim 1, characterized in that: The first text sentence matrix is ​​divided according to the extracted subject words to obtain a plurality of second text sentence matrices, including: The speech texts in the first text sentence matrix are classified according to the extracted subject words, the speech texts belonging to the same subject word are divided into the same category, and a second text sentence matrix is ​​generated.

5. The method for extracting in-vehicle speech abstracts according to claim 1, characterized in that: The position features of the second text sentence matrix are marked to obtain a sentence position matrix, including: Determining the acquisition time of the speech text in the second text sentence matrix, Determine to sort the speech text once according to the acquisition time; When the acquisition time is the same, determining the word order position relationship of the speech text in the second text sentence matrix; The speech texts acquired at the same time are sorted again according to the word order position relationship to obtain a sentence position matrix.

6. The method for extracting in-vehicle speech abstracts according to claim 1, characterized in that: The sentence position matrix is ​​weighted and the weight of each sentence position is determined, including: The sentence position matrix is ​​weighted to obtain the sentence position matrix: Normalize the sentence position matrix to get the matrix Y i ; According to the matrix Y i Get the horizontal information entropy E i and longitudinal information entropy E j ; According to the horizontal information entropy E i and longitudinal information entropy E j Determine the weight of each sentence position.

7. The method for extracting in-vehicle speech abstracts according to claim 6, characterized in that: The matrix Y i Each element Y ij Calculated according to the following formula: Among them, x ij is the sentence position matrix Xi and Y ij The element corresponding to the position, min(x i ) is the minimum value, max(x i ) is the maximum value.

8. The method for extracting in-vehicle speech abstracts according to claim 7, characterized in that: According to the horizontal information entropy E i and longitudinal information entropy E j Determine the weight of each sentence position, including: According to the horizontal information entropy E i Determine the horizontal weight w of each sentence position i , calculate the lateral weight w according to the following formula i ; According to the longitudinal information entropy E j Determine the vertical weight w of each sentence position j , calculate the longitudinal weight w according to the following formula j ; 9. The method for extracting in-vehicle speech abstracts according to claim 8, characterized in that: The comprehensive score of each word vector is determined according to the TF-IDF value of each word vector and the weight of each sentence position, and the summary is extracted according to the comprehensive score, including: Determine the speech text to which each word vector belongs, and determine the weight of the corresponding sentence position according to the speech text to which it belongs; The comprehensive score of each word vector is determined based on the TF-IDF value of the word vector and the weight of the corresponding sentence position. The comprehensive score is calculated according to the following formula: C score =TF×IDF×w i ×w j ; A summary is extracted from the word vector according to the comprehensive score.

10. A system for extracting in-vehicle speech abstracts, used for applying the method for extracting in-vehicle speech abstracts as claimed in any one of claims 1 to 9, characterized in that: The system comprises: A voice text acquisition module, the acquisition module is used to acquire the voice text in the car; A sentence matrix division module, the sentence matrix division module is used to divide the speech text according to the vehicle journey to obtain a plurality of first text sentence matrices; and extract a subject word from each first text sentence matrix, and divide the first text sentence matrix according to the extracted subject words to obtain a plurality of second text sentence matrices; A TF-IDF value determination module, the TF-IDF value determination module is used to segment the second text sentence matrix to obtain word vector text, and determine the TF-IDF value of each word vector according to the word vector text; A sentence weight determination module, the sentence weight determination module is used to mark the position features of the second text sentence matrix to obtain a sentence position matrix; and to weight the sentence position matrix and determine the weight of each sentence position; The summary extraction module is used to determine the comprehensive score of each word vector according to the TF-IDF value of each word vector and the weight of each sentence position, and perform summary extraction according to the comprehensive score.

Images