New word updating method, speech recognition device and computer readable storage medium

Abstract

The embodiment of the present invention provides a new word updating method, a speech recognition device and a computer-readable storage medium, which extracts new words to be verified from received sentences, then sends the new words to a preset entry search engine, and receives the search results sent by the entry search engine, thereby determining to add the new words to the pronunciation dictionary based on the search results without the need for manual subjective judgment, thereby improving the updating efficiency of the new words in the pronunciation dictionary and improving the updating accuracy of the new words in the pronunciation dictionary.

Description

Technical Field

[0001] The embodiments of the present invention relate to, but are not limited to, the field of speech recognition, and specifically, relate to, but are not limited to, a new word updating method, a speech recognition device, and a computer-readable storage medium. Background Art

[0002] In the related art, in the Automatic Speech Recognition (ASR) system, it is very important whether the entries in the pronunciation dictionary are complete and reasonable. When introducing foreign corpus (i.e. foreign sentences) to construct new words, currently only new words that do not exist in the pronunciation dictionary can be subjectively identified from the corpus manually and added to the pronunciation dictionary. This approach has the problems of low updating efficiency and accuracy, which is not conducive to improving the generalization performance of ASR. Summary of the invention

[0003] The embodiments of the present invention provide a new word updating method, a speech recognition device and a computer-readable storage medium, which mainly solve the technical problem of: solving the problem that the related technology manually identifies new words that do not exist in the pronunciation dictionary from the corpus and adds them to the pronunciation dictionary, resulting in low updating efficiency and accuracy.

[0004] To solve the above technical problems, an embodiment of the present invention provides a new word updating method, comprising: extracting new words to be verified from received sentences; sending the new words to a preset entry search engine, and receiving search results sent by the entry search engine, wherein the search results include the existence of the new words on the entry search engine; and determining to add the new words to the pronunciation dictionary based on the search results.

[0005] An embodiment of the present invention also provides a speech recognition device, including a processor, a memory and a communication bus; the communication bus is used to realize the connection and communication between the processor and the memory; the processor is used to execute one or more computer programs stored in the memory to implement the steps of the new word updating method as described above.

[0006] An embodiment of the present invention further provides a computer-readable storage medium, which stores one or more computer programs. The one or more computer programs can be executed by one or more processors to implement the steps of the new word updating method as described above.

[0007] According to the new word updating method, speech recognition device and computer-readable storage medium provided by the embodiments of the present invention, new words to be verified are extracted from the received sentences, and then the new words are sent to a preset entry search engine, and search results sent by the entry search engine are received, and the search results include the existence of the new words on the entry search engine; thereby, the new words are added to the pronunciation dictionary based on the search results, without the need for manual subjective judgment, which can improve the updating efficiency of the new words in the pronunciation dictionary, and improve the updating accuracy of the new words in the pronunciation dictionary, and is more conducive to improving the generalization performance of ASR.

[0008] Other features and corresponding beneficial effects of the present invention are described in the latter part of the specification, and it should be understood that at least part of the beneficial effects become obvious from the description in the specification of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A schematic diagram of a new word updating method flow chart provided in the first embodiment of the present invention;

[0010] Figure 2 Schematic diagram of the process of extracting new words to be verified provided in the first embodiment of the present invention Figure 1 ;

[0011] Figure 3 Schematic diagram of the process of extracting new words to be verified provided in the first embodiment of the present invention Figure 2 ;

[0012] Figure 4 A schematic diagram of a process for determining whether to add a new word to a pronunciation dictionary provided in Embodiment 1 of the present invention Figure 1 ;

[0013] Figure 5 A schematic diagram of a process for determining whether to add a new word to a pronunciation dictionary provided in Embodiment 1 of the present invention Figure 2 ;

[0014] Figure 6 A schematic diagram of the structure of a new word updating device provided in the second embodiment of the present invention;

[0015] Figure 7 A schematic diagram of a new word updating method flow chart provided in the second embodiment of the present invention;

[0016] Figure 8 This is a schematic diagram of the structure of a speech recognition device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the following is a further detailed description of the embodiments of the present invention through specific implementation methods combined with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] Embodiment 1:

[0019] In order to improve the efficiency and accuracy of updating new words in the pronunciation dictionary in the ASR system, the new word updating method provided in this embodiment utilizes an entry search engine. The new words to be verified extracted from the corpus (i.e., sentences) can be searched using the entry search engine, thereby determining whether to add the new words to the pronunciation dictionary based on the search results without the need for manual subjective judgment. This can not only improve the updating efficiency of new words in the pronunciation dictionary, but also improve the updating accuracy of new words in the pronunciation dictionary, and is more conducive to improving the generalization performance of ASR.

[0020] For ease of understanding, this embodiment is described below in conjunction with Figure 1 The new word update method shown is an example for explanation, see Figure 1 As shown, it includes but is not limited to:

[0021] S101: extracting new words to be verified from the received sentences.

[0022] It should be understood that the received statement in this embodiment may be a statement issued by a user, and the user may issue the statement during the use of the ASR function, or during the testing or debugging of the ASR function. The received statement in this embodiment may also be a statement from other devices.

[0023] In this embodiment, the new words to be verified can be extracted from the received sentence automatically by the speech recognition device, or can be manually selected, or a combination of manual selection and automatic execution by the speech recognition device. For ease of understanding, this embodiment is described below using the method of automatically extracting the new words to be verified by the speech recognition device as an example.

[0024] Extraction method 1, please refer to Figure 2 As shown, the steps of extracting new words to be verified from the received sentences include but are not limited to:

[0025] S201: Perform word segmentation processing on the sentence.

[0026] In this embodiment, a word segmentation dictionary can be used to segment the sentence. The word segmentation dictionary may include but is not limited to at least one of the word segmentation dictionaries in the ASR system and the word segmentation dictionaries in natural language processing, that is, at least one of the word segmentation methods in the ASR system and the word segmentation methods in natural language processing can be adopted but not limited to.

[0027] For example, assuming that the received sentence is "nearby restaurants with good reviews", it can be segmented to obtain "nearby", "reviews", "good", and "restaurant".

[0028] S202: Among the words obtained through the word segmentation process, words that do not exist in the current pronunciation dictionary are regarded as new words to be verified.

[0029] In this step, each word obtained by the word segmentation process in the previous step is compared with the existing words in the current pronunciation dictionary, so that the words that do not exist in the current pronunciation dictionary are selected as new words to be verified.

[0030] For example, assuming that the words "nearby", "good", and "restaurant" already exist in the current pronunciation dictionary, and the word "evaluation" does not exist in the current pronunciation dictionary, the word "evaluation" can be used as a new word to be verified.

[0031] For extraction method 2, please refer to Figure 3 As shown, the steps of extracting new words to be verified from the received sentences include but are not limited to:

[0032] S301: Perform word segmentation processing on the sentence.

[0033] In this embodiment, a word segmentation dictionary can be used to segment the sentence. The word segmentation dictionary may include but is not limited to at least one of the word segmentation dictionaries in the ASR system and the word segmentation dictionaries in natural language processing, that is, at least one of the word segmentation methods in the ASR system and the word segmentation methods in natural language processing can be adopted but not limited to.

[0034] For example, assuming that the received sentence is "navigation to the People's Bank of China within five kilometers", it can be processed by word segmentation to obtain "navigation", "within five kilometers", and "People's Bank of China".

[0035] S302: Among the words obtained through word segmentation, words that do not exist in the current pronunciation dictionary are used as pre-checked words.

[0036] In this step, each word obtained by the word segmentation process in the previous step is compared with the existing words in the current pronunciation dictionary, so that the words that do not exist in the current pronunciation dictionary are selected as pre-checked words.

[0037] For example, assuming that the word "navigation" already exists in the current pronunciation dictionary, and the words "within five kilometers" and "People's Bank of China" do not exist in the current pronunciation dictionary, the words "within five kilometers" and "People's Bank of China" can be used as pre-checked words.

[0038] S303: performing word segmentation processing on the words to be checked again, and taking at least one of the words obtained after the word segmentation processing that is not in the current pronunciation dictionary as a new word to be checked.

[0039] In this step, when the pre-checked words are segmented again, the segmentation method used can be the same as the segmentation method used in S301 above, or different from the segmentation method used in S301 above, which can be flexibly set according to application requirements. The words obtained by segmenting the pre-checked words are compared with the existing words in the current pronunciation dictionary, so as to select the pre-checked words that are not fully present in the current pronunciation dictionary as the pre-checked words.

[0040] For example, suppose the pre-verification word "within five kilometers" is segmented again to obtain "five kilometers" and "within", and suppose "five kilometers" exists in the current pronunciation dictionary and "within" does not exist in the current pronunciation dictionary, then the pre-verification word "within five kilometers" will be used as a new word to be verified.

[0041] Assume that the pre-verification word "People's Bank of China" is segmented again to obtain "China", "name", and "bank". Assume that the three words "China", "name", and "bank" all exist in the current pronunciation dictionary, and the pre-verification word "People's Bank of China" is not treated as a new word to be verified.

[0042] It should be understood that the method of extracting new words to be verified from the received sentences in this embodiment is not limited to the above two examples. Other equivalent extraction methods can also be used according to application requirements, which will not be repeated here.

[0043] S102: Send the new word to a preset entry search engine, and receive the search result sent by the entry search engine, where the search result includes the existence of the new word on the entry search engine.

[0044] In this embodiment, the term search engine may include one or more terms. When one term search engine is used, the efficiency is relatively higher. When multiple terms are used, the coverage is more comprehensive, and the judgment based on the search results of each term search engine is more accurate.

[0045] It should be understood that the term search engine in this embodiment may include various mature term search engines, such as including but not limited to various mature encyclopedia term search engines (such as but not limited to Baidu Encyclopedia, Wikipedia, Sogou Encyclopedia); it may also include a newly created term search engine for term search.

[0046] In this embodiment, sending the new word to a preset term search engine and receiving the search results sent by the term search engine may include but is not limited to: for each term search engine, constructing an engine search request including the new word, and sending the constructed engine search requests to their respective corresponding term search engines.

[0047] For example, assuming that the entry search engines include Baidu Encyclopedia, Wikipedia, and Sogou Encyclopedia, engine search requests containing new words can be constructed for Baidu Encyclopedia, Wikipedia, and Sogou Encyclopedia respectively, and sent to Baidu Encyclopedia, Wikipedia, and Sogou Encyclopedia respectively.

[0048] In this embodiment, the engine search request may be constructed by, but not limited to, a crawler tool and sent to a corresponding entry search engine.

[0049] S103: Adding new words to the pronunciation dictionary based on the search results.

[0050] In this embodiment, whether the new word to be verified needs to be added to the pronunciation dictionary can be determined directly based on the existence of the new word to be verified in the search results of each term search engine on each term search engine.

[0051] In this embodiment, the method of adding the new word to the pronunciation dictionary based on the search results may also include but is not limited to any one of the following two methods.

[0052] Method 1, see Figure 4 As shown, determining whether to add a new word to the pronunciation dictionary based on the search results includes but is not limited to:

[0053] S401: Counting the first number of search results for new words in the search results sent by each term search engine.

[0054] S402: When the first number obtained by counting is greater than or equal to a set first threshold, determine to add the new word to the pronunciation dictionary.

[0055] For example, if only Baidu Encyclopedia is used as a search engine for a term in an example, the first threshold value can be set to 1, and then the search results returned by Baidu Encyclopedia are counted as new words on the Baidu Encyclopedia search engine. If so, the first number is 1, which is greater than or equal to the first threshold value 1, and the new word is determined to be added to the pronunciation dictionary. Otherwise, the first number is 0, which is less than or equal to the first threshold value 1, and the new word is not added to the pronunciation dictionary.

[0056] For another example, assuming that Baidu Encyclopedia, Wikipedia, and Sogou Encyclopedia are used as three entry search engines in an example, the first threshold can be set to 2, and then the search results returned by Baidu Encyclopedia, Wikipedia, and Sogou Encyclopedia are counted to see whether they are new words on their respective entry search engines. If so, the first number is increased by 1. When the first number counted reaches 2, it is greater than or equal to the first threshold 2, and it is determined that the new word is added to the pronunciation dictionary. When the first number counted is ultimately less than or equal to the first threshold 2, it is not added to the pronunciation dictionary.

[0057] Optionally, in some application examples, when the first number is less than a first threshold, the new word may be invalidated, or the new word may be put into a manual error correction library for manual secondary judgment.

[0058] Method 2: In this method, in order to further improve the accuracy of judgment, the presence of the neighboring words of the new word in the sentence on the entry search engine can also be combined to determine whether to add the new word to the pronunciation dictionary. This method can avoid the situation where the judgment result is unstable due to relying solely on the presence of the new word itself on each entry search engine. For example, "the game against Shandong West Kings" is divided into "against Shandong", "East and West", "Wang Team", "of", and "game", among which the word "East and West" itself has a high presence, but it is caused by the mis-segmentation of the text, so considering the presence of the contextual neighboring words of the new word in the sentence text on the entry search engine can further improve the accuracy of judgment.

[0059] In this method, before determining whether to add a new word to the pronunciation dictionary based on the search results, the method further includes:

[0060] At least one neighboring word of the new word in the sentence is sent to a preset entry search engine, and a neighboring word search result sent by the entry search engine is received; the neighboring word is a word of a preset type.

[0061] In this embodiment, the preset type of neighboring words can be other types of neighboring words excluding preposition types; it can also be flexibly set to other specified types of words according to the application scenario requirements.

[0062] In this method, the specific proximity relationship between the new word and at least one neighboring word of the preset type in the sentence can be flexibly set. For example:

[0063] In one example, when the new word is not the first or last word in the sentence, the previous (or two or more as required) neighboring word of the new word in the sentence and the next (or two or more as required) neighboring word of the new word in the sentence can be selected. When the new word is the first word in the sentence, the next (or two or more as required) neighboring word of the new word in the sentence can be selected; when the new word is the last word in the sentence, the previous (or two or more as required) neighboring word of the new word in the sentence can be selected.

[0064] In this method, whether to add a new word to the pronunciation dictionary is determined based on the search results. Figure 5 As shown, it includes but is not limited to:

[0065] S501: Counting a second number of search results for new words on the search results sent by each entry search engine; and counting a third number of search results for neighboring words on the search results sent by each entry search engine;

[0066] S502: When the sum of the second number and the third number is greater than or equal to a set second threshold, determine to add the new word to the pronunciation dictionary.

[0067] Optionally, in this embodiment, when the above S101 extracts the new word to be verified from the received sentence using the extraction method 1 of the above example, when it is determined to add the new word to the pronunciation dictionary based on the search results, before adding the new word to the pronunciation dictionary, it may also include:

[0068] The new word is further segmented, and when at least one of the words obtained after the segmentation does not exist in the current pronunciation dictionary, the new word is added to the pronunciation dictionary to further improve the accuracy of the judgment.

[0069] Assuming that the new word "People's Bank of China" is determined to be added to the pronunciation dictionary based on the search results, it is segmented again to obtain "China", "name" and "bank". Assuming that the three words "China", "name" and "bank" all exist in the current pronunciation dictionary, the new word "People's Bank of China" will not be added to the pronunciation dictionary; it can be invalidated or put into the manual error correction library for manual secondary judgment; the specific settings can be flexibly based on needs.

[0070] Optionally, in this embodiment, in order to ensure that the search results are from the term search engine, after receiving the search results sent by each term search engine and before counting them, it is also possible to optionally determine whether the format of the received search results meets the format of the corresponding term search engine, and if so, count them; otherwise, do not count them. This ensures the accuracy of the source of the search results.

[0071] It can be seen that the new word updating method provided in this embodiment can use the entry search engine to extract new words to be verified from the corpus (i.e., sentences), and then determine whether to add the new words to the pronunciation dictionary based on the existence of the new words in the entry search engine, without the need for manual subjective judgment. This can not only improve the updating efficiency of new words in the pronunciation dictionary, but also improve the updating accuracy of new words in the pronunciation dictionary, and is more conducive to improving the generalization performance of ASR.

[0072] Embodiment 2:

[0073] This embodiment provides a new word updating device, which can be set on a speech recognition device (for example, but not limited to a server). Figure 6 As shown, it includes but is not limited to:

[0074] Word segmentation module 601, corpus acquisition module 602, search module 603, identification module 604 and pronunciation dictionary 605;

[0075] The corpus acquisition module 602 is used to acquire corpus, that is, acquire sentences. The acquisition method is shown in the above examples and will not be repeated here.

[0076] The word segmentation module 601 is used to extract new words to be verified from the sentences obtained by the corpus acquisition module 602. The extraction method is shown in the above examples and will not be repeated here.

[0077] The search module 603 is used to send new words to a preset entry search engine and receive search results sent by the entry search engine; the specific sending and receiving methods are shown in the above examples and will not be repeated here; the search module 603 can be implemented by but not limited to a crawler tool.

[0078] The determination module 604 is used to add new words to the pronunciation dictionary 605 based on the search results. The specific determination method is shown in the above examples and will not be repeated here.

[0079] It should be understood that the function of at least one of the above modules can be implemented by, but not limited to, a processor in the speech recognition device.

[0080] For ease of understanding, this embodiment is described below with reference to a specific application scenario as an example. Figure 7As shown, the new word updating method in this example includes but is not limited to:

[0081] S701: The acquired new text corpus (ie, sentence) is segmented through a word segmenter to obtain a word sequence set.

[0082] For example, assuming that the sentence obtained is "the company chooses the Microsoft cloud service platform", the word sequence set of "company", "select", "Microsoft", "cloud service", and "platform" is obtained through the word segmentation.

[0083] S702: Compare each word in the word sequence set with the words in the current pronunciation dictionary to obtain a new word set to be verified.

[0084] For example, assuming that among the words “company”, “select”, “Microsoft”, “cloud service” and “platform”, “cloud service” does not exist in the current pronunciation dictionary, it is added as a new word to be verified into the new word set to be verified.

[0085] S703: For the new word to be checked, construct search engine search requests for each entry corresponding to the new word through a crawler tool, and access these engines sequentially or in parallel.

[0086] For example, the crawler tool is used to construct the search request URL address corresponding to the new word "cloud service" in each encyclopedia entry search engine (here, Baidu Encyclopedia, Wikipedia, and Sogou Encyclopedia are used). For example, for the Baidu Encyclopedia engine, the constructed search request URL address is:

[0087] https: / / baike.baidu.com / item / Cloud Service

[0088] These URL requests are then sent to the corresponding encyclopedia entry search engine servers, and responses, i.e., search results, are returned by the servers.

[0089] S704: Count the existence of the new word on each term search engine based on the search results returned by each term search engine, and count the existence of at least one neighboring word of the new word in the sentence on each term search engine.

[0090] In this example, the HTML format of the search results returned by each term search engine is parsed. If the format meets the specific term format of the corresponding term engine (for example, whether the HTML format returned by Baidu Encyclopedia contains a specific DIV tag named 'j-shareAbstract', rather than just returning the new term), it is then determined whether the term exists on the term search engine. If so, the existence degree (i.e., the second quantity) of the new term is increased by 1, otherwise, the existence degree is decreased by 1 or remains unchanged. Assuming that in this example, "cloud service" has the correct term format returned in Baidu Encyclopedia, Wikipedia, and Sogou Encyclopedia, the existence degree, i.e., the second quantity, is 3.

[0091] In this example, for the new word "cloud service" in the above example, it is also necessary to obtain the existence of the two words "Microsoft" and "platform" on each entry search engine. The specific acquisition process is shown in the above examples and will not be repeated here. In this example, it is assumed that the two words "Microsoft" and "platform" also exist on Baidu Encyclopedia, Wikipedia, and Sogou Encyclopedia, that is, the third quantity statistical result is 3+3=6.

[0092] The comprehensive existence score of “Cloud Service” (i.e. the sum of the second and third quantities) = “Microsoft”’s own existence score + “Cloud Service”’s own existence score + “Platform”’s own existence score = 3+3+3=9.

[0093] Finally, the discriminator determines whether the new word can be added to the pronunciation dictionary based on the comprehensive existence score threshold of the new word (i.e., the second threshold). The second threshold can be adaptively obtained by statistical methods to obtain a suitable value, or a simple value can be set based on experience, such as 5. Obviously, in this example, the new word "cloud service" is judged as a reasonable new term that can be added to the pronunciation dictionary of the ASR system.

[0094] S705: Determine whether the sum of the second quantity and the third quantity is greater than or equal to a second threshold value, if not, go to S706; if yes, go to S707.

[0095] S706: The new word is put into the wrong word library for manual error correction.

[0096] S707: Use the current pronunciation dictionary (other segmentation methods may also be used) to segment the new word again.

[0097] S708: Determine whether all words obtained after the new word segmentation exist in the current pronunciation dictionary; if so, go to S709; otherwise, go to S710.

[0098] S709: Abandon the new word.

[0099] S710: No, include the new word into the pronunciation dictionary.

[0100] For example, for the new word "cloud service", it is segmented to obtain the two words "cloud" and "service". Assuming that "cloud" and "service" both exist in the current pronunciation dictionary, the new word "cloud service" will be invalidated; otherwise, it will be added to the pronunciation dictionary.

[0101] Embodiment three:

[0102] This embodiment also provides a speech recognition device, which can be but is not limited to a server, see Figure 8 As shown, it includes a processor 801, a memory 802 and a communication bus 803, wherein:

[0103] The communication bus 803 is used to realize the connection and communication between the processor 801 and the memory 802;

[0104] The processor 801 is configured to execute one or more computer programs stored in the memory 802 to implement at least one step in the new word updating method in each of the above embodiments.

[0105] The present embodiment also provides a computer-readable storage medium, which includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer.

[0106] The computer-readable storage medium in this embodiment can be used to store one or more computer programs, and the one or more computer programs stored therein can be executed by a processor to implement at least one step of the new word updating method exemplified in the above-mentioned embodiments.

[0107] This embodiment also provides a computer program (or computer software), which can be distributed on a computer-readable medium and executed by a computing device to implement at least one step of the new word updating method in the above-mentioned embodiment; and in some cases, at least one step of the new word updating method different from the above-mentioned embodiment examples can be adopted.

[0108] It should be understood that, in some cases, at least one step shown or described may be performed in an order different from that described in the above embodiments.

[0109] This embodiment further provides a computer program product, including a computer readable device, on which the computer program as shown above is stored. In this embodiment, the computer readable device may include the computer readable storage medium as shown above.

[0110] It can be seen that those skilled in the art should understand that all or some of the steps, systems, and functional modules / units in the above disclosed methods can be implemented as software (which can be implemented with computer program code executable by a computing device), firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component can have multiple functions, or a function or step can be performed by several physical components in cooperation. Some or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit.

[0111] In addition, it is well known to those skilled in the art that communication media generally contain computer readable instructions, data structures, computer program modules or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery media. Therefore, the present invention is not limited to any specific hardware and software combination.

[0112] The above contents are further detailed descriptions of the embodiments of the present invention in combination with specific implementation methods, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A new word updating method, comprising: Extracting new words to be verified from the received sentences; When the preset entry search engine includes multiple entry search engines, constructing an engine search request including the new word for each entry search engine, sending the constructed engine search request to the corresponding entry search engine, and receiving search results sent by the entry search engine, wherein the search results include the existence of the new word on the entry search engine; Sending at least one neighboring word of the new word in the sentence to the preset entry search engine, and receiving the neighboring word search result sent by the entry search engine; The adjacent words are words of a preset type; adding the new word to a pronunciation dictionary based on the search results; The adding the new word to the pronunciation dictionary based on the search results comprises: Counting a second number of search results for the new word on the term search engine among the search results sent by each term search engine; and counting a third number of neighboring word search results for the neighboring word existing on the word search engine among the neighboring word search results sent by each of the word search engines; When the sum of the second number and the third number is greater than or equal to a set second threshold, it is determined to add the new word to the pronunciation dictionary.

2. The new word updating method according to claim 1, characterized in that: The adding the new word to the pronunciation dictionary based on the search results comprises: Counting the first number of search results for the new word on the term search engine among the search results sent by each term search engine; When the first number is greater than or equal to a set first threshold, the new word is added to the pronunciation dictionary.

3. The new word updating method according to claim 1 or 2, characterized in that: The step of extracting a new word to be verified from the received sentence includes: Performing word segmentation processing on the sentence; Among the words obtained by the word segmentation process, the words that do not exist in the current pronunciation dictionary are regarded as the new words to be verified.

4. The new word updating method as claimed in claim 3, characterized in that: When it is determined based on the search results that the new word is to be added to the pronunciation dictionary, before adding the new word to the pronunciation dictionary, the method further comprises: The new word is further segmented, and when at least one of the words obtained after the segmentation does not exist in the current pronunciation dictionary, the new word is added to the pronunciation dictionary.

5. The new word updating method according to claim 1 or 2, characterized in that: The step of extracting a new word to be verified from the received sentence includes: Among the words obtained by the word segmentation process, the words that are not in the current pronunciation dictionary are used as pre-checked words; The pre-checked words are further segmented, and at least one of the words obtained after the segmentation that is not in the current pronunciation dictionary is used as a new word to be checked.

6. A speech recognition device, characterized in that: including a processor, a memory and a communication bus; The communication bus is used to realize the connection and communication between the processor and the memory; The processor is used to execute one or more computer programs stored in the memory to implement the steps of the new word updating method according to any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more computer programs, and the one or more computer programs can be executed by one or more processors to implement the steps of the new word updating method according to any one of claims 1 to 5.

Images