CN111354339B - Vocabulary phoneme list construction method, device, equipment and storage medium - Google Patents

Abstract

The invention discloses a vocabulary phoneme table construction method, device, equipment and storage medium. The method includes: selecting several vocabulary conversion phonetic symbols tools, and using the several vocabulary conversion phonetic symbols tools to mark phonetic symbols for the words to be marked, and obtain the phonetic symbols. Describe several phonetic symbols of the vocabulary to be marked; based on the voting strategy, select the winning phonetic symbol from the several phonetic symbols as the target phonetic symbol; convert the target phonetic symbol into phonemes to generate a vocabulary phoneme table. As a result, multiple vocabulary-to-phonetic notation tools are used to annotate phonetic symbols for the words to be tagged, and the target phonetic symbols are determined based on the voting strategy, which improves the quality of the vocabulary phoneme table and improves the efficiency of constructing the vocabulary phoneme table.

Description

Vocabulary phoneme table construction method, device, equipment and storage medium

Technical field

The present invention relates to the technical field of speech recognition, and in particular to a method, device, equipment and storage medium for constructing a vocabulary phoneme table.

Background technique

With the development of computer technology, more and more technologies (big data, distributed, blockchain, artificial intelligence, etc.) are applied in the financial field. The traditional financial industry is gradually transforming into financial technology (Fintech). However, due to the financial The security and real-time requirements of the industry also put forward higher requirements for technology.

The lexicon table (lexicon table) is a key part of building a hybrid speech recognition system. Generally, in order to improve speech recognition performance, speech recognition systems need to convert words into more fine-grained phonemes. The current way to build a vocabulary phoneme table is to manually label or use the open source lexicon table, but manual labeling is time-consuming and labor-intensive. The quality of the vocabulary phoneme table built by the open source lexicon table cannot be guaranteed, and in most cases it lacks professional vocabulary in professional fields.

Contents of the invention

The present invention provides a vocabulary phoneme table construction method, device, equipment and storage medium, aiming to improve the quality of the vocabulary phoneme table and improve the construction efficiency of the vocabulary phoneme table.

In order to achieve the above object, the present invention provides a method for constructing a vocabulary phoneme table, which method includes:

Select several vocabulary-to-phonetic notation tools, and use the several vocabulary-to-phonetic notation tools to respectively annotate phonetic symbols for the words to be tagged, and obtain several phonetic symbols of the words to be tagged;

Based on the voting strategy, select the winning phonetic symbol from the several phonetic symbols as the target phonetic symbol;

Convert the target phonetic symbols into phonemes to generate a vocabulary phoneme table.

Preferably, the voting strategy is used to determine the winning phonetic symbol based on the number of votes for the phonetic symbol;

The step of selecting the winning phonetic symbol as the target phonetic symbol from the several phonetic symbols based on the voting strategy includes:

Each of the plurality of phonetic symbols is assigned to an original ticket;

If there are identical phonetic symbols among the several phonetic symbols, the original votes for the same phonetic symbols will be combined, and the combined phonetic symbols will be marked as candidate phonetic symbols, and the number of votes for the candidate phonetic symbols will be counted;

The candidate phonetic symbols are sorted according to the number of votes, the winning phonetic symbol is determined according to the sorting result, and the winning phonetic symbol is used as the target phonetic symbol.

Preferably, after the step of assigning each of the plurality of phonetic symbols to an original ticket, the step further includes:

If there are no identical phonetic symbols among the several phonetic symbols, it is determined that there is no winning phonetic symbol among the several phonetic symbols;

Mark the corresponding words to be labeled as ambiguous words, and transfer the ambiguous words into the ambiguous word pool;

The ambiguous words in the ambiguous word pool are phonetically marked using several backup word-to-phonetic notation tools until the winning phonetic notation of the ambiguous word is obtained.

Preferably, before the step of selecting a winning phonetic symbol from the several phonetic symbols as the target phonetic symbol based on the voting strategy, the step further includes:

Determine whether the word to be tagged is a polysyllabic word based on the number of phonetic symbols annotated by the word-to-phonetic symbol conversion tool:

If the number of phonetic symbols marked by one or more of the vocabulary conversion tools is greater than 1, it is determined that the vocabulary to be marked is a polysyllabic word;

If the vocabulary to be tagged is a polysyllabic word, the polysyllabic word is split into several sub-vocabularies to be tagged;

Respectively store the several sub-vocabularies to be marked in association with the corresponding phonetic symbols, and after obtaining the several phonetic symbols of the sub-vocabulary to be marked, perform the following steps: based on the voting strategy, select the winning phonetic symbol from the several phonetic symbols as the target phonetic symbol .

Preferably, the step of selecting several vocabulary-to-phonetic notation tools, and using the several vocabulary-to-phonetic notation tools to respectively annotate phonetic symbols for the words to be annotated, and obtaining the several phonetic symbols of the words to be annotated, further includes:

The several phonetic symbols are normalized to obtain several phonetic symbols with the same format, so as to select the winning phonetic symbol from the several phonetic symbols with the same format.

Preferably, the step of converting the target phonetic symbols into phonemes and generating a vocabulary phoneme table includes:

Based on the phoneme format, the target phonetic symbols are converted into phonemes, and a vocabulary phoneme table is generated according to the phonemes of the vocabulary to be annotated.

Preferably, the step of converting the target phonetic symbols into phonemes and generating a vocabulary phoneme table further includes:

Receive a vocabulary phoneme table update request, and obtain the target update vocabulary and target update operation from the vocabulary phoneme table update request;

Based on the target update operation, a corresponding update operation is performed on the phonemes of the target vocabulary.

In addition, to achieve the above object, the present invention also provides a device for constructing a vocabulary phoneme table. The device for constructing a vocabulary phoneme table includes:

A selection module, used to select several vocabulary-to-phonetic symbols conversion tools, and use the several vocabulary-to-phonetic symbols conversion tools to respectively mark phonetic symbols for the words to be marked, and obtain several phonetic symbols of the words to be marked;

A voting module, used to select the winning phonetic symbol from the several phonetic symbols as the target phonetic symbol based on the voting strategy;

A conversion module, used to convert the target phonetic symbols into phonemes and generate a vocabulary phoneme table.

In addition, to achieve the above object, the present invention also provides a vocabulary phoneme table construction device. The vocabulary phoneme table construction device includes a processor, a memory and a vocabulary phoneme table construction program stored in the memory. The vocabulary phoneme table construction program When the construction program is run by the processor, the steps of the vocabulary phoneme table construction method as described above are implemented.

In addition, in order to achieve the above object, the present invention also provides a computer storage medium. A vocabulary phoneme table construction program is stored on the computer storage medium. When the vocabulary phoneme table construction program is run by the processor, the vocabulary phoneme table is implemented as described above. Steps to build a method.

Compared with the existing technology, the present invention provides a method, device, equipment and storage medium for constructing a vocabulary phoneme table. The method includes: selecting a number of vocabulary conversion tools for phonetic notation, and using the several vocabulary conversion tools for phonetic notation to convert words to be annotated. Mark phonetic symbols to obtain several phonetic symbols of the vocabulary to be marked; based on the voting strategy, select the winning phonetic symbol from the several phonetic symbols as the target phonetic symbol; convert the target phonetic symbols into phonemes to generate a vocabulary phoneme table. Therefore, multiple vocabulary-to-phonetic notation tools are used to annotate phonetic symbols for the words to be annotated, and the target phonetic symbols are determined based on the voting strategy, which overcomes the energy consumed by manual annotation and avoids the lack of certain vocabulary due to manual annotation or open source lexicon tables, such as Uncommon vocabulary and professional problems lead to inaccurate vocabulary phoneme tables, which improves the quality of vocabulary phoneme tables and improves the efficiency of building vocabulary phoneme tables.

Description of drawings

Figure 1 is a schematic diagram of the hardware structure of a vocabulary phoneme table construction device involved in various embodiments of the present invention;

Figure 2 is a schematic flow chart of the first embodiment of the vocabulary phoneme table construction method of the present invention;

Figure 3 is a schematic flowchart of the second embodiment of the vocabulary phoneme table construction method of the present invention;

Figure 4 is a schematic flow chart of the third embodiment of the vocabulary phoneme table construction method of the present invention;

Figure 5 is a schematic diagram of the functional modules of the first embodiment of the vocabulary phoneme table construction device of the present invention.

The realization of the purpose, functional features and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present invention and are not intended to limit the present invention.

The vocabulary phoneme table construction device mainly involved in the embodiment of the present invention refers to a network connection device capable of realizing network connection. The vocabulary phoneme table construction device may be a server, a cloud platform, etc.

Referring to FIG. 1 , FIG. 1 is a schematic diagram of the hardware structure of a device for constructing a vocabulary phoneme table according to various embodiments of the present invention. In the embodiment of the present invention, the vocabulary phoneme table construction device may include a processor 1001 (such as a Central Processing Unit, CPU), a communication bus 1002, an input port 1003, an output port 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize connection communication between these components; the input port 1003 is used for data input; the output port 1004 is used for data output. The memory 1005 can be a high-speed RAM memory or a stable memory (non-volatile). memory), such as disk memory. The memory 1005 may optionally be a storage device independent of the aforementioned processor 1001. Those skilled in the art can understand that the hardware structure shown in Figure 1 does not limit the present invention, and may include more or fewer components than shown, or combine certain components, or arrange different components.

Continuing to refer to FIG. 1 , the memory 1005 as a readable storage medium in FIG. 1 may include an operating system, a network communication module, an application module, and a vocabulary phoneme table construction program. In Figure 1, the network communication module is mainly used to connect to the server and perform data communication with the server; and the processor 1001 can call the vocabulary phoneme table construction program stored in the memory 1005, and execute the vocabulary phoneme table construction method provided by the embodiment of the present invention. .

The embodiment of the present invention provides a method for constructing a vocabulary phoneme table.

Referring to Figure 2, Figure 2 is a schematic flow chart of a first embodiment of a method for constructing a vocabulary phoneme table of the present invention.

In this embodiment, the vocabulary phoneme table construction method is applied to a vocabulary phoneme table construction device, and the method includes:

Step S101, select several vocabulary-to-phonetic notation tools, use the several vocabulary-to-phonetic notation tools to annotate phonetic symbols for the words to be tagged, and obtain several phonetic symbols of the words to be tagged;

The vocabulary phoneme table is a key part of building a hybrid speech recognition system. Usually, in order to ensure the recognition effect, the speech recognition system needs to convert words into more fine-grained phonemes. There are millions of words, but only a few hundred well-designed phonemes. By modeling coarser-grained phonemes, the speech recognition system can significantly reduce the search space and improve the recognition effect. In this embodiment, words to be tagged are prepared in advance, and the words to be tagged include Chinese words, English words, Japanese words, French words, etc. The words to be labeled include commonly used words, rare words, professional words, etc.

Further, select several of the vocabulary-to-phonetic notation tools, and the number of the vocabulary-to-phonetic notation tools is greater than or equal to 3. The tools for converting words to phonetic symbols include, but are not limited to, tools for converting Chinese words to pinyin and tools for converting English words to American pronunciation phonetic symbols.

The word-to-phonetic symbol conversion tool can convert words into corresponding phonetic symbols. For example, for the word "patent", it can be converted into the phonetic notation [zhuan1 li4]; for patent, it can be converted into the American phonetic notation ['peitnt].

The forms of phonetic symbols annotated by different word-to-phonetic notation tools may be different. Therefore, after obtaining several phonetic symbols of the vocabulary to be labeled, the several phonetic symbols are normalized to obtain several phonetic symbols with the same format. Specifically, the format of the normalized phonetic symbols is first set, and then the several phonetic symbols are normalized into the format. For example, for Chinese vocabulary, the format can be set to "Pinyin + Tone", and the tones are represented by the numbers 1 (one tone), 2 (second tone), 3 (third tone), 4 (fourth tone), and 5 (soft tone). For example, if the phonetic symbols marked by a word-to-phonetic notation tool use "ˉ, ˊ, ˇ, ˋ" to represent tones, it will be converted into tones represented by numbers. For example, for the word "ma", if the phonetic symbol marked by a word conversion tool is [má], it will be normalized to [ma2].

It can be understood that after the words to be marked are marked with phonetic symbols through several of the vocabulary-to-phonetic notation tools, each word has a corresponding number of phonetic symbols, and the number of vocabulary phonetic symbols is equal to the number of the vocabulary-to-phonetic notation tools. correspond. For example, if the words to be tagged are tagged using 5 vocabulary-to-phonetic symbol conversion tools, and if the words to be tagged are monosyllabic words, then each word to be tagged has a corresponding 5 phonetic symbols; if the words to be tagged are If it is a polysyllabic word, the number of phonetic symbols corresponding to each word to be tagged is greater than 5.

Step S102: Based on the voting strategy, select the winning phonetic symbol from the several phonetic symbols as the target phonetic symbol;

It is understandable that the accuracy of each vocabulary-to-phonetic notation tool is not 100%, so the phonetic symbols obtained by each vocabulary-to-phonetic notation tool may be different after phonetic notation of the words to be annotated. This embodiment uses a voting strategy to select the correct phonetic symbol from several phonetic symbols of the vocabulary to be marked.

Specifically, the step S102: based on the voting strategy, the step of selecting the winning phonetic symbol as the target phonetic symbol from the several phonetic symbols includes:

Step S102a: Assign each of the several phonetic symbols to an original ticket;

Each phonetic symbol mentioned above is subject to voting, and each person can receive an original vote. In this embodiment, the original tickets are virtual tickets, and the same number of original tickets are assigned to each phonetic symbol. To facilitate vote counting, this embodiment sets the number of original votes to one.

Step S102b: If there are identical phonetic symbols among the several phonetic symbols, combine the original votes for the same phonetic symbols, mark the combined phonetic symbols as candidate phonetic symbols, and count the number of votes for the candidate phonetic symbols;

Generally speaking, considering that the accuracy of word-to-phonetic notation tools is not 100%, the same phonetic symbol may exist if the same word is annotated by different word-to-phonetic notation tools. Compare the phonetic symbols marked by the phonetic symbol conversion tool for each word, filter out the same phonetic symbols, and combine the original tickets for the same phonetic symbols. In this embodiment, scripts, statistical tools, etc. can be used to aggregate the original tickets.

Further, the combined phonetic symbols are marked as candidate phonetic symbols, and the votes of the candidate phonetic symbols are counted and saved for subsequent selection of the winning phonetic symbol.

Step S102c: Sort the candidate phonetic symbols according to the number of votes, determine the winning phonetic symbol according to the sorting result, and use the winning phonetic symbol as the target phonetic symbol.

After the votes are combined, the number of votes for each of the candidate phonetic symbols is calculated, and the corresponding candidate phonetic symbols are sorted according to the sorting rules based on the number of votes. If the sorting rule is forward sorting, that is, sorted by the number of votes from high to low, the candidate phonetic symbol ranked first will be determined as the winning phonetic symbol; if the sorting rule is reverse sorting, that is, sorted by the number of votes from low to high Sorting, then the candidate phonetic symbol ranked first from the bottom is determined as the winning phonetic symbol, and the winning phonetic symbol is used as the target phonetic symbol.

For example, for the word "中", if the number of the phonetic symbols is 5, and these 5 phonetic symbols are [zhong1], [zhong1], [zong1], [zhong1], [zong1], then they are respectively These 5 phonetic symbols are assigned to one original ticket, and 3 [zhong1] and 2 [zong1] are combined. The [zhong1] and [zong1] are the candidate phonetic symbols. If forward sorting is used as the sorting rule, [zhong1] and [zong1] will be sorted forward according to the number of votes. Since [zhong1] has 3 votes and [zong1] has 2 votes, the sorting result is: [zhong1]>[zong1 ]; [zhong1] ranks first, so [zhong1] is determined to be the winning phonetic symbol.

If the several phonetic symbols of the vocabulary to be marked include one or more phonetic symbols that can be combined, and one or more phonetic symbols that cannot be combined, then the number of phonetic symbols after the combination is definitely greater than the phonetic symbols that cannot be combined. number of votes, so one or more phonetic symbols that cannot be combined can be ignored.

In addition, the minimum number of votes for the winning phonetic symbol can also be set according to the number of vocabulary conversion tools. For example, if the number of vocabulary conversion tools for phonetic notation is 7, then the minimum number of votes for the winning phonetic notation is set to 5. In this way, the accuracy of the vocabulary phoneme table can be further improved.

Further, the step of assigning each of the plurality of phonetic symbols to an original ticket also includes:

Step S102a1: If there are no identical phonetic symbols among the several phonetic symbols, determine that there is no winning phonetic symbol among the several phonetic symbols;

If there are no identical phonetic symbols, it means that the phonetic symbols marked by each word-to-phonetic notation tool are different. Therefore, it is difficult to determine which word-to-phonetic symbol tool has the correct phonetic symbol, so it is determined that there is no winning phonetic symbol.

Step S102a2: Mark the corresponding words to be labeled as ambiguous words, and transfer the ambiguous words into the ambiguous word pool;

If there is no winning phonetic symbol, it is difficult to ensure the accuracy of the phonetic symbol labeling of the word. Therefore, the words to be labeled that do not have the winning phonetic symbol are marked as ambiguous words, and the ambiguous words are transferred to the ambiguous word pool.

Step S102a3: Use several backup vocabulary conversion tools to annotate phonetic symbols for the ambiguous words in the ambiguous word pool until the winning phonetic symbol of the ambiguous word is obtained.

For the ambiguous vocabulary, select several backup vocabulary conversion tools to re-annotate phonetic symbols for the vocabulary in the ambiguous vocabulary pool. At least one of the several backup word-to-phonetic notation tools is different from the previously used word-to-phonetic notation tool. The words in the ambiguous vocabulary pool can be manually labeled. Actual tests have proven that the ambiguous words generated by the technical solution of this embodiment account for less than 1% of the words to be labeled. Therefore, even if the ambiguous words are manually labeled, it can be completed without spending too much manpower.

It can be understood that in other embodiments, different voting strategies can be set, and the winning phonetic symbol can be obtained according to the voting strategy. For example, first group the several phonetic symbols, put the same phonetic symbols into the same group, and obtain multiple groups of phonetic symbols; then count the number of phonetic symbols in each group of phonetic symbols, and sort the candidate phonetic symbols according to the number of votes. The result was determined to be the winner of phonetic symbols.

Step S103: Convert the target phonetic symbols into phonemes and generate a vocabulary phoneme table.

Specifically, based on the phoneme format, the target phonetic symbols are converted into phonemes, and the phonemes of the vocabulary to be annotated are saved into a vocabulary phoneme table. The phoneme format is determined in advance, and the target phonetic symbols are converted into phonemes based on the phoneme format. The phoneme format includes phoneme connection methods, phoneme arrangement orders, etc.; wherein the phoneme connection methods include dot-symbol connection, space connection, etc. For example, for the word "Mandarin" to be annotated, the corresponding target phonetic symbol is [pu1tong1hua4], and the corresponding phoneme is represented as [p u1t ong1 h ua4].

After obtaining the phonemes of all words in the vocabulary to be tagged, the phonemes are saved and the phoneme table of the words is generated.

In this embodiment, through the above solution, several vocabulary-to-phonetic notation tools are selected, and the several vocabulary-to-phonetic notation tools are used to annotate phonetic symbols for the words to be tagged, and several phonetic symbols of the words to be tagged are obtained; based on the voting strategy, from the Select the winning phonetic symbol from several phonetic symbols as the target phonetic symbol; convert the target phonetic symbol into phonemes to generate a vocabulary phoneme table. As a result, multiple vocabulary-to-phonetic notation tools are used to annotate phonetic symbols for the words to be tagged, and the target phonetic symbols are determined based on the voting strategy, which improves the quality of the vocabulary phoneme table and improves the efficiency of constructing the vocabulary phoneme table.

As shown in Figure 3, the second embodiment of the present invention proposes a method for constructing a vocabulary phoneme table. Based on the first embodiment shown in Figure 2, the winning phonetic symbol is selected from the several phonetic symbols based on the voting strategy. As a step before the target phonetic symbols, it also includes:

Step S1011: Determine whether the word to be marked is a polysyllabic word based on the number of phonetic symbols marked by the word-to-phonetic symbol conversion tool;

The polysyllabic words refer to words with two or more phonetic symbols, and polysyllabic words are the abbreviation of polysyllabic words with different sounds and the same character. Specifically, whether the word to be marked is a polysyllabic word is determined based on the number of phonetic symbols marked by the phonetic symbol conversion tool for each word.

Step S1012: If the number of phonetic symbols marked by one or more of the vocabulary conversion tools is greater than 1, determine that the vocabulary to be marked is a polysyllabic word;

In this embodiment, whether the word to be labeled is a polysyllabic word is determined based on the number of phonetic symbols. If the number of phonetic symbols marked by one or more of the word-to-phonetic symbol tools is greater than 1, it is determined that the word to be marked is a polysyllabic word. For example, for the word "Chaoyang", there are two phonetic symbols [zhao1 yang2] and [chao2 yang2]. If the word is annotated through multiple vocabulary-to-phonetic notation tools, at least one word must be converted to phonetic notation, taking into account the error rate and circumstances. The tool can mark these two phonetic symbols.

Step S1013: If the vocabulary to be tagged is a polysyllabic word, split the vocabulary to be tagged into several sub-vocabularies to be tagged.

In this embodiment, the number of sub-vocabularies to be labeled is the same as the number of phonetic symbols. For example, "Chaoyang" can be divided into two sub-vocabularies.

Further, in the process of dividing the words into sub-vocabularies, the words to be annotated themselves are mapped to the most commonly used phonetic symbols, and other phonetic symbols other than the most commonly used phonetic symbols are mapped in the form of "vocabulary + suffix". The suffix can be letters, numbers, etc. In this embodiment, the most commonly used phonetic symbols refer to the most frequently used phonetic symbols in sentences containing the vocabulary in each sentence library. For example, for the polyphonic word "wan", the phonetic symbols are [wan4] and [mo4] respectively. Among them, the most commonly used phonetic symbol is [wan4], which is mapped to [万wan4] and [万_2mo4].

Step S1014: Store the plurality of sub-vocabularies to be labeled in association with the corresponding phonetic symbols, and after obtaining the plurality of phonetic symbols of the sub-vocabulary to be labeled, perform the step of: based on the voting strategy, select the winning phonetic symbol from the several phonetic symbols. as the target phonetic symbol.

The several sub-vocabularies to be tagged are stored in association with the corresponding phonetic symbols, and the mapping results of each sub-vocabulary to be tagged are respectively marked. In this way, several phonetic symbols of the sub-vocabulary to be labeled are obtained.

After obtaining several phonetic symbols of the sub-vocabulary to be labeled, step S102 is performed: based on the voting strategy, select the winning phonetic symbol from the several phonetic symbols as the target phonetic symbol.

In this embodiment, through the above solution, several vocabulary-to-phonetic notation tools are selected, and the several vocabulary-to-phonetic notation tools are used to annotate phonetic symbols for the words to be tagged, and several phonetic symbols of the words to be tagged are obtained; based on the voting strategy, from the Select the winning phonetic symbol from several phonetic symbols as the target phonetic symbol; convert the target phonetic symbol into phonemes to generate a vocabulary phoneme table. As a result, multiple vocabulary-to-phonetic notation tools are used to annotate phonetic symbols for the words to be tagged, and the target phonetic symbols are determined based on the voting strategy, which improves the quality of the vocabulary phoneme table and improves the efficiency of constructing the vocabulary phoneme table.

As shown in Figure 4, the third embodiment of the present invention proposes a method for constructing a vocabulary phoneme table. Based on the first and second embodiments shown in Figures 2 and 3, the target phonetic symbols are converted into Phonemes, the steps to generate the vocabulary phoneme table also include:

Step S104, receive a vocabulary phoneme table update request, and obtain the target update vocabulary and target update operation from the vocabulary phoneme table update request;

After the construction of the vocabulary phoneme table is completed, in order to obtain a more complete and accurate vocabulary phoneme table, operations such as modification, addition, and deletion of the vocabulary phoneme table need to be performed.

Specifically, a vocabulary phoneme table update request is received, the vocabulary phoneme table update request includes an update operation, and the update operation includes modification, deletion, and addition. The vocabulary to be updated and the corresponding target update operation are obtained from the vocabulary phoneme table update request, and the target update operation includes one of modification, addition, and deletion.

Step S105: Based on the target update operation, perform a corresponding update operation on the phonemes of the target vocabulary.

The target vocabulary to be updated and the target update operation corresponding to the target vocabulary are obtained from the vocabulary phoneme table update request. If the target update operation is to modify, then further obtain the modified phoneme, and then replace the phoneme of the target vocabulary in the vocabulary phoneme table with the modified phoneme; if the target update operation is to add, then obtain the new phoneme that needs to be The added target vocabulary and the phonemes of the target vocabulary are then saved to the vocabulary phoneme table.

In this embodiment, the newly added or modified phonemes may be manually annotated phonemes, or may be phonemes obtained based on the solution of the first embodiment of the present invention.

In this embodiment, through the above solution, several vocabulary-to-phonetic notation tools are selected, and the several vocabulary-to-phonetic notation tools are used to annotate phonetic symbols for the words to be tagged, and several phonetic symbols of the words to be tagged are obtained; based on the voting strategy, from the Select the winning phonetic symbol from several phonetic symbols as the target phonetic symbol; convert the target phonetic symbol into phonemes, generate a vocabulary phoneme table, receive a vocabulary phoneme table update request, and obtain the target update vocabulary and target update operation from the vocabulary phoneme table update request. ; Based on the target update operation, perform a corresponding update operation on the phonemes of the target vocabulary. As a result, multiple vocabulary-to-phonetic notation tools are used to annotate phonetic symbols for the words to be tagged, and the target phonetic symbols are determined based on the voting strategy, which improves the quality of the vocabulary phoneme table and improves the efficiency of constructing the vocabulary phoneme table. It also enables fast updates.

In addition, this embodiment also provides a device for constructing a vocabulary phoneme table. Referring to Figure 5, Figure 5 is a schematic diagram of the functional modules of the first embodiment of the vocabulary phoneme table construction device of the present invention.

In this embodiment, the vocabulary phoneme table construction device is a virtual device, which is stored in the memory 1005 of the vocabulary phoneme table construction device shown in Figure 1 to realize all functions of the vocabulary phoneme table construction program: for selecting several vocabulary words A phonetic symbol conversion tool, which uses the several vocabulary conversion phonetic symbols to mark phonetic symbols for words to be marked, and obtains several phonetic symbols of the words to be marked; and is used to select a winning phonetic symbol from the several phonetic symbols based on a voting strategy as the phonetic symbol. Target phonetic symbols; used to convert the target phonetic symbols into phonemes and generate a vocabulary phoneme table.

Specifically, the vocabulary phoneme construction device includes:

The selection module 10 is used to select several vocabulary conversion phonetic notation tools, and use the several vocabulary conversion phonetic notation tools to respectively mark phonetic symbols for the vocabulary to be marked, and obtain several phonetic symbols of the vocabulary to be marked;

The voting module 20 is used to select the winning phonetic symbol from the several phonetic symbols as the target phonetic symbol based on the voting strategy;

The conversion module 30 is used to convert the target phonetic symbols into phonemes and generate a vocabulary phoneme table.

Further, the voting module includes:

An assigning unit, configured to assign each of the plurality of phonetic symbols to an original ticket;

A ticket summarizing unit is used to, if there are identical phonetic symbols among the plurality of phonetic symbols, combine the original tickets for the same phonetic symbols, mark the combined phonetic symbols as candidate phonetic symbols, and count the number of votes for the candidate phonetic symbols;

A determining unit, configured to sort the candidate phonetic symbols according to the number of votes, and determine the winning phonetic symbol according to the sorting result.

Further, the granting unit also includes:

A determination subunit, used to determine that there is no winning phonetic symbol among the several phonetic symbols if there are no identical phonetic symbols among the several phonetic symbols;

The marking subunit is used to mark the corresponding words to be labeled as ambiguous words, and transfer the ambiguous words into the ambiguous word pool;

Obtaining a subunit for using several backup vocabulary conversion tools to annotate phonetic symbols for ambiguous words in the ambiguous word pool until the winning phonetic symbol of the ambiguous words is obtained.

Further, the voting module also includes:

A judging unit, configured to judge whether the word to be labeled is a polysyllabic word based on the number of phonetic symbols labeled by each word conversion tool:

A determination unit, configured to determine that the vocabulary to be marked is a polyphonetic word if the number of phonetic symbols marked by one or more of the vocabulary conversion tools is greater than 1;

A splitting unit, used to split the polysyllabic word into several sub-vocabularies to be tagged if the word to be tagged is a polysyllabic word;

The storage unit is used to store the plurality of sub-vocabularies to be labeled in association with the corresponding phonetic symbols, and after obtaining the several phonetic symbols of the sub-vocabulary to be labeled, perform the step of: selecting from the several phonetic symbols based on the voting strategy. The winning phonetic symbol is used as the target phonetic symbol.

Further, the location selection module also includes:

The several phonetic symbols are normalized to obtain several phonetic symbols with the same format, so as to select the winning phonetic symbol from the several phonetic symbols with the same format.

Further, the conversion module also includes:

Based on the phoneme format, the target phonetic symbols are converted into phonemes, and a vocabulary phoneme table is generated according to the phonemes of the vocabulary to be annotated.

Further, the conversion module also includes:

An acquisition unit configured to receive a vocabulary phoneme table update request, and obtain a target update vocabulary and a target update operation from the vocabulary phoneme table update request;

An update unit, configured to perform a corresponding update operation on the phonemes of the target vocabulary based on the target update operation.

In addition, embodiments of the present invention further provide a computer storage medium, which stores a vocabulary phoneme table construction program. When the vocabulary phoneme table construction program is run by a processor, the vocabulary phoneme table construction method is implemented as described above. The steps will not be repeated here.

Compared with the prior art, the present invention proposes a method, device, equipment and storage medium for constructing a vocabulary phoneme table. The method includes: selecting a number of vocabulary-to-phonetic symbol conversion tools, and using the several vocabulary-to-phonetic symbol conversion tools to convert words to be annotated. Vocabulary annotation phonetic symbols are obtained to obtain several phonetic symbols of the vocabulary to be annotated; based on the voting strategy, the winning phonetic symbols are selected from the several phonetic symbols as the target phonetic symbols; the target phonetic symbols are converted into phonemes to generate a vocabulary phoneme table. As a result, multiple vocabulary-to-phonetic notation tools are used to annotate phonetic symbols for the words to be tagged, and the target phonetic symbols are determined based on the voting strategy, which improves the quality of the vocabulary phoneme table and improves the efficiency of constructing the vocabulary phoneme table.

It should be noted that, as used herein, the terms "include", "comprising" or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or system that includes a list of elements not only includes those elements, but It also includes other elements not expressly listed or that are inherent to the process, method, article or system. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

The above serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product that is essentially or contributes to the existing technology. The computer software product is stored in a storage medium (such as ROM/RAM) as mentioned above. , magnetic disk, optical disk), including several requests to cause a terminal device to execute the method described in various embodiments of the present invention.

The above are only preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or process transformation made using the description and drawings of the present invention may be directly or indirectly applied in other related technical fields. , are all similarly included in the scope of patent protection of the present invention.

Claims (9)

1. A method of vocabulary phonemic table construction, the method comprising:

selecting a plurality of vocabulary phonetic symbol conversion tools, and respectively marking phonetic symbols for the vocabulary to be marked by the plurality of vocabulary phonetic symbol conversion tools to obtain a plurality of phonetic symbols of the vocabulary to be marked;

selecting winning phonetic symbols from the several phonetic symbols as target phonetic symbols based on voting strategy;

converting the target phonetic symbol into a phoneme to generate a vocabulary phoneme list;

before the step of selecting the winning phonetic symbol from the plurality of phonetic symbols as the target phonetic symbol based on the voting strategy, the method further comprises the following steps:

judging whether the vocabulary to be annotated is a polyphonic word or not according to the number of phonetic symbols annotated by each vocabulary to be annotated:

if the number of phonetic symbols marked by one or more vocabulary phonetic symbol conversion tools is greater than 1, judging that the vocabulary to be marked is a multi-phonetic-symbol;

if the vocabulary to be annotated is a polyphone, splitting the polyphone into a plurality of sub-vocabularies to be annotated;

respectively storing the plurality of sub-vocabularies to be marked and corresponding phonetic symbols in an associated manner, and executing the steps after obtaining the plurality of phonetic symbols of the sub-vocabularies to be marked: and selecting a winning phonetic symbol from the several phonetic symbols as a target phonetic symbol based on a voting strategy.

2. The method of claim 1, wherein the voting strategy is used to determine winning phonetic symbols based on the number of votes for the phonetic symbols;

the step of selecting the winning phonetic symbol from the plurality of phonetic symbols as the target phonetic symbol based on the voting strategy comprises the following steps:

each phonetic symbol in the plurality of phonetic symbols is respectively endowed with an original ticket;

if the same phonetic symbols exist in the plurality of phonetic symbols, the original ticket of the same phonetic symbol is combined, the phonetic symbols after the ticket combination are marked as candidate phonetic symbols, and the ticket number of the candidate phonetic symbols is counted;

and sorting the candidate phonetic symbols according to the ticket number, determining winning phonetic symbols according to the sorting result, and taking the winning phonetic symbols as target phonetic symbols.

3. The method of claim 2, wherein after the step of assigning each of the plurality of phonetic symbols to an original ticket, further comprising:

if the same phonetic symbols do not exist in the plurality of phonetic symbols, judging that the winning phonetic symbols do not exist in the plurality of phonetic symbols;

marking the corresponding vocabulary to be annotated as ambiguous vocabulary, and converting the ambiguous vocabulary into an ambiguous vocabulary pool;

and annotating the ambiguous vocabulary in the ambiguous vocabulary pool by a plurality of spare vocabulary phonetic transcription tools until the winning phonetic transcription of the ambiguous vocabulary is obtained.

4. The method of claim 1, wherein the selecting a plurality of vocabulary phonetic transcription tools, the plurality of vocabulary phonetic transcription tools respectively annotate phonetic transcriptions of the vocabulary to be annotated, and after the step of obtaining the plurality of phonetic transcriptions of the vocabulary to be annotated, further comprises:

normalizing the plurality of phonetic symbols to obtain a plurality of phonetic symbols with consistent formats, so as to select winning phonetic symbols from the plurality of phonetic symbols with consistent formats.

5. The method of claim 1, wherein the step of converting the target phonetic symbol to phonemes and generating a vocabulary phonemic list comprises:

and converting the target phonetic symbol into a phoneme based on a phoneme format, and generating a vocabulary phoneme list according to the phonemes of the vocabulary to be annotated.

6. The method of claim 1, wherein the step of converting the target phonetic symbol to phonemes and generating a vocabulary phonemic list further comprises:

receiving a vocabulary phonemic table updating request, and acquiring a target updating vocabulary and a target updating operation from the vocabulary phonemic table updating request;

and based on the target updating operation, executing corresponding updating operation on the phonemes of the target updating vocabulary.

7. A vocabulary phoneme list construction device, characterized in that the vocabulary phoneme list construction device comprises:

the selection module is used for selecting a plurality of vocabulary phonetic symbol conversion tools, and the vocabulary phonetic symbol conversion tools are used for respectively marking phonetic symbols for the vocabulary to be marked to obtain a plurality of phonetic symbols of the vocabulary to be marked;

the voting module is used for selecting winning phonetic symbols from the plurality of phonetic symbols as target phonetic symbols based on a voting strategy;

the conversion module is used for converting the target phonetic symbol into a phoneme to generate a vocabulary phoneme list;

the voting module further comprises:

the judging unit is used for judging whether the vocabulary to be marked is a polyphone according to the number of phonetic symbols marked by each vocabulary phonetic symbol converting tool:

the judging unit is used for judging that the vocabulary to be marked is a polyphone if the number of phonetic symbols marked by one or more vocabulary phonetic symbol conversion tools is larger than 1;

the splitting unit is used for splitting the multi-sound word into a plurality of sub-words to be annotated if the words to be annotated are multi-sound words;

the storage unit is used for respectively storing the plurality of sub-vocabularies to be marked and the corresponding phonetic symbols in an associated way, and executing the steps after obtaining the plurality of phonetic symbols of the sub-vocabularies to be marked: and selecting a winning phonetic symbol from the several phonetic symbols as a target phonetic symbol based on a voting strategy.

8. A vocabulary phonemic building device comprising a processor, a memory and a vocabulary phonemic building program stored in the memory, which vocabulary phonemic building program, when executed by the processor, implements the steps of the vocabulary phonemic building method according to any one of claims 1-6.

9. A computer storage medium having stored thereon a vocabulary phoneme list construction program which when executed by a processor performs the steps of the vocabulary phoneme list construction method of any one of claims 1to 6.