A speech recognition method for power dispatching

Abstract

The present invention discloses a speech recognition method for the field of power dispatching, which is used to recognize voice commands and dialogues in the process of power dispatching. Power dispatching usually involves a large number of power professional terms and place names, lines, power stations, equipment names with specific numbers and even names of people that are unique to each power department. Therefore, the error rate of commonly used speech recognition methods is relatively high. The present invention proposes the difference frequency principle, automatically establishes a local power difference frequency special vocabulary, matches pinyin and text with the difference frequency special vocabulary as the center, and adopts an arbitrary position conversion mechanism. Through these steps, a high-accuracy recognition effect can be obtained without a special voice library for model training, especially the special vocabulary of the local power department can be recognized.

Description

Technical Field

[0001] The present invention relates to the technical field of speech recognition, and in particular to a speech recognition method for the field of electric power dispatching. Background Art

[0002] In the process of power dispatching, using speech recognition technology to recognize the dispatcher's voice instructions and dialogues can improve efficiency and further realize dispatching automation. However, power dispatching usually involves a large number of power professional terms and the names of places, lines, power stations, equipment names with specific numbers and even names of people that are unique to each power department. Due to the existence of a large number of homophones in the language, the current speech recognition technology often recognizes these relatively uncommon special words as other common ordinary words, so the error rate is relatively high and it is difficult to meet the needs of the power profession. Therefore, the main reason for the above problems is that the current speech recognition technology is based on frequency priority matching. When the voice is converted to pinyin, it will give priority to matching those common words and popular words that appear more frequently.

[0003] Some literature proposes adding dedicated vocabulary, but there are still three problems:

[0004] 1. The special vocabulary of the local power department will not appear in the general vocabulary database. The power departments in each region need to manually establish it according to the needs of their own departments. The dispatchers need to screen the special vocabulary one by one from a large amount of local text materials. In addition, these special vocabulary must be constantly updated, replaced and statistically refreshed, which is time-consuming and laborious. The dispatching department usually has heavy responsibilities and heavy tasks and it is difficult to split its time.

[0005] 2. Add special words and common popular words together in the vocabulary. Special words appear less frequently, but popular words will still be given priority when their pinyin is the same.

[0006] 3. Currently, the matching starts with the first letter of a pinyin string and converts it into text one by one. Due to the interference of noise and the existence of homophones, some pinyins of important special words in the back may be combined with the previous pinyins in advance and converted into other phrases, resulting in errors. For example, the knife switch of XXX line was originally required to be closed, but it turned out to be closed on other lines, causing serious consequences. On the contrary, errors in common words are relatively acceptable. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and propose a speech recognition method for the field of power dispatching, which can recognize special vocabulary in power dispatching at a relatively low cost, especially the special vocabulary of the local power department, such as local place names, lines, power stations, specific equipment names and even personal names.

[0008] To achieve the above-mentioned purpose, the technical solution provided by the present invention is: a speech recognition method for the field of power dispatching, comprising:

[0009] Level priority matching: after converting speech to pinyin, a pinyin string consisting of letters and tones is obtained, which is named A. In the process of converting A into text, it is preferentially matched with the pinyin of the first-level sub-library vocabulary of the difference frequency special vocabulary. If the match is successful, part of the pinyin of A is converted into text. If the match is unsuccessful, the next level is considered, until the last level sub-library. The matching is to find the similarity between a part of the pinyin of A and the correct pinyin of a Chinese phrase or character, which is also referred to as the matching of pinyin and text or vocabulary. The difference frequency special vocabulary is composed of hierarchical special vocabulary and its pinyin. The level of the vocabulary is determined by the difference between its two frequencies, that is, the higher the frequency of occurrence in special materials, the higher the level of the vocabulary, and the higher the frequency of occurrence in general materials, the lower the level of the vocabulary. The vocabulary refers to Chinese phrases, and a Chinese phrase includes at least 2 Chinese characters. The special vocabulary includes local special vocabulary and professional terms. The local special vocabulary refers to vocabulary used only in the local machine, local area network, a specific region, group or department. Special vocabulary of the same level constitutes a sub-library. The highest level sub-library is the first level sub-library, and the following are from the second level to the lowest level sub-library.

[0010] Frequency priority matching: After the level priority matching is completed, the remaining pinyin of A is matched with the pinyin of common vocabulary. Non-specialized vocabulary with high frequency appearing in common data is matched first. Finally, the remaining pinyin is matched with the pinyin of a single Chinese character.

[0011] Further, the step of obtaining a pinyin string composed of letters and tones includes:

[0012] S1. Authority identification: Perform voiceprint recognition on the wake-up voice input by the user. If the user has the authority to use, proceed to step S2, otherwise, exit;

[0013] S2, speech noise reduction: perform speech noise reduction on user speech to reduce the interference of noise on recognition;

[0014] S3, voice to pinyin conversion: perform voice analysis and recognition on the voice after noise reduction in step S2, and obtain the whole sentence pinyin string corresponding to the voice.

[0015] Further, the permission identification includes the following steps:

[0016] S11, obtaining a wake-up voice;

[0017] S12, extracting the Fbank energy features of each frame in the speech obtained in step S11;

[0018] S13, input the Fbank energy features into a pre-trained DNN model to obtain a high-level feature expression based on deep learning;

[0019] S14. Perform cosine similarity matching between the high-level feature expression obtained in step S13 and the user voice feature bound to the user login account; if the cosine similarity is lower than or equal to the set value, it is determined that the user does not match the user bound to the login account and has no use permission; if it is higher than the set value, it is determined that the user has use permission.

[0020] Further, the construction of the difference frequency special word library includes the following steps:

[0021] S1. Collect vocabulary of specialized data, including local power dispatch logs, work tickets, equipment records and call texts, and search for power professional academic articles on the Internet;

[0022] S2. Clean and segment the vocabulary of the special data collected in the previous step to obtain a vocabulary list, and then count the word frequency of the vocabulary list to obtain a special word frequency dictionary; wherein, word frequency = the number of times the word is repeated × the length of the word / the total number of words in all the data, and the special word frequency dictionary contains all the words and their word frequencies that appear in the special data;

[0023] S3, collect vocabulary from general data, including the People's Daily corpus and news from three major websites, Sina, Sohu, and NetEase, and perform vocabulary frequency statistics to obtain a general word frequency dictionary, which includes all the vocabulary and word frequencies that appear in the general data;

[0024] S4, perform a difference frequency operation on each word in the special word frequency dictionary to obtain a difference frequency special word library, wherein the difference frequency operation is:

[0025] Difference frequency The special word frequency of a word - k × its general word frequency, where k is a fixed coefficient. The special word frequency and general word frequency are obtained from the special word frequency dictionary and the general word frequency dictionary respectively. The word frequency that cannot be found in the dictionary is counted as 0;

[0026] S5. The top 25% of the words with the difference frequency value are stored in the first-level sub-bank, the words between 26% and 50% are stored in the second-level sub-bank, the words between 51% and 75% are stored in the third-level sub-bank, and the rest of the words with a difference frequency value greater than 0 are stored in the fourth level. Words with a difference frequency value less than or equal to 0 are discarded;

[0027] S6. Associate each word in the frequency difference special word library with its corresponding pinyin, and finally form a frequency difference special word library for power dispatching; the frequency difference special word library is composed of first, second, third and fourth level sub-libraries, and the higher the frequency difference value, the higher the word is in the queue of the sub-library.

[0028] Furthermore, the level priority matching step includes:

[0029] Reverse word selection: select the pinyin of the word with the highest difference frequency value from the words that have not been matched in the first-level sub-library, and set its name as B. If all the words in the first-level sub-library have been matched, then move on to the next-level sub-library;

[0030] Convert pinyin at any position: Search for a substring C similar to B in A. If B matches C successfully, convert C into the corresponding Chinese phrase. If there are multiple substrings similar to B in A, repeat the above operation. The substring C can be located at any position in A.

[0031] Furthermore, it also includes: when the recognized voice is a conversation, subject word determination, subject word queue sorting and subject word matching are performed before level priority matching, as follows:

[0032] Determine the subject word: count the words repeated in the previous text and their repetition times; if the repeated word is a frequency-delayed special word, add the word to the subject word queue, otherwise discard it;

[0033] Sorting of topic word queue: Assume that n conversation sentences have been recognized from the start of this speech recognition to the current sentence to be recognized, and the current sentence to be recognized is numbered as sentence n+1, then the topic value of a repeated word is:

[0034] ;

[0035] In the formula, i and j are the words repeated in the i-th and j-th sentences, and the ellipsis indicates other repeated sentences; i, j < n; G is the level of the sub-library of the difference frequency special word library to which the word belongs, and its value is an integer from 1 to 4; calculate the topic values ​​of all the topic words in the first n sentences, and then arrange them from large to small according to the topic values ​​to obtain the topic word queue;

[0036] Subject word matching: Match A with the subject word queue, starting from the first subject word in the queue. If the match is successful, part of the pinyin of A will be converted into text. If the match is unsuccessful, consider the next subject word until the last subject word in the queue.

[0037] Further, the matching comprises the following steps:

[0038] Calculate the phoneme edit distance: refers to the minimum number of phoneme edit operations required to convert one pinyin string into another. The phoneme refers to the initial consonant or final of the pinyin. The permitted edit operations include: inserting an initial consonant or final, deleting an initial consonant or final, replacing one initial consonant or final with another. The replacement between ambiguous sounds is only counted as 0.5 times. The above operations do not include tones.

[0039] If the matching word is a special word, then when the phoneme edit distance is less than a given threshold, the phoneme edit distance and a match success signal are output, otherwise a match failure signal is given;

[0040] If the match is a common vocabulary, the phoneme edit distance is output.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] 1. According to the method of the present invention, the speech recognition system can automatically distinguish between common words and special words in the field of power dispatching, especially those special words of local power departments, so that there is no need for power departments in various regions to manually build databases. The system will store these special words in a hierarchical difference frequency special word library and continuously refresh, update and replace them, thereby saving a lot of time and energy of dispatchers.

[0043] 2. The level priority matching of the method of the present invention focuses on and prioritizes the special vocabulary for power dispatching, reducing the errors caused by the prior art speech recognition method due to the priority of popular general vocabulary, thereby improving the accuracy of speech recognition in power dispatching, reducing errors in the recognition of dispatching instructions, reducing power dispatching accidents, improving dispatching efficiency, and better realizing intelligent dispatching. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of the method of the present invention.

[0045] Figure 2 It is a flow chart of the voiceprint recognition process.

[0046] Figure 3 It is a flow chart of the speech noise reduction process. DETAILED DESCRIPTION

[0047] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0048] See also Figure 1 As shown, this embodiment discloses a speech recognition method for the field of power dispatching, and the specific implementation steps are as follows:

[0049] S1. Authority identification. Perform voiceprint recognition on the wake-up voice input by the user. If the user has the authority to use, proceed to step S2. Otherwise, exit the voice recognition system.

[0050] S2, voice noise reduction. Perform voice noise reduction on the user's voice to reduce the interference of noise on recognition;

[0051] S3. Convert speech to pinyin. Use a well-known deep learning speech recognition algorithm to perform speech analysis and recognition on the noise-reduced speech obtained in S2, and obtain the full-sentence pinyin corresponding to the speech. For example, when the user's command speech is: "Insert the 35th and 36th earthing switches on the Yuegang Xiangshi Line at Yuetang Station", through the conversion in step S3, [tou2 ru4 yue4 tang2 zhan4 yue4 gang1 xiang1 shi2 xian4 san1 wu3 jie1 di4 dao1 zha2 he2 san1 liu4 jie1 di4 dao1 zha2] will be obtained, which is called pinyin string A;

[0052] S4. Use the level-priority hierarchical matching method to perform Chinese text matching on the full-sentence pinyin. For example, Yuetang Station, Yuegang Xiangshi Line, and earthing switch are all special terms. The differential frequency value ranking is: Yuetang Station (level 1) > Yuegang Xiangshi Line (level 2) > earthing switch (level 3). 1) Reverse word extraction: Extract words one by one from the first-level sub-library in descending order of differential frequency value, and for each word extracted, check whether there is a matching substring in pinyin string A. The current matching method is to extract pinyin from string A and search in the vocabulary library. The method of this patent is the opposite, so it is called reverse word extraction; 2) Any-position conversion: The current method starts converting text from the first letter. This method is different. The substring can be converted at any position in string A. If the matching gap is greater than the given threshold, it is abandoned, and the next word is taken until the pinyin of Yuetang Station "yue4 tang2 zhan4" can match the corresponding part in pinyin string A, so that pinyin string A becomes [tou2 ru4 Yuetang Station yue4 gang1 xiang1 shi2 xian4 san1 wu3 jie1 di4 dao1 zha2 he2 san1 liu4 jie1 di4 dao1 zha2]. Here, reverse word extraction and any-position conversion are specifically designed for differential-frequency special terms and are different from the currently well-known methods. Similarly, the remaining special terms in string A will be converted later: [tou2 ru4 Yuetang Station Yuegang Xiangshi Line san1 wu3 earthing switch he2 san1 liu4 earthing switch];

[0053] S5. Match pinyin and general vocabulary with frequency priority. When all the special terms in string A are converted, match the general vocabulary according to the well-known frequency-priority method: in the order from front to back, take tou2 ru4, look up the general dictionary, and get "投入" (insert), and string A becomes: [Insert Yuetang Station Yuegang Xiangshi Line san1 wu3 earthing switch he2 san1 liu4 earthing switch];

[0054] S6, the remaining pinyin matches a single Chinese character to obtain the entire sentence text [put into Yuetang Station Yuegang Xiangshi Line 35 grounding knife switch and 36 grounding knife switch];

[0055] S7, output the whole sentence;

[0056] S8, refresh vocabulary frequency, difference frequency value and sorting; refresh the difference frequency value of difference frequency vocabulary Yuetang Station and Yuegang Xiangshi Line and update its sorting in the library. The difference frequency vocabulary that does not appear does not need to be frequently refreshed;

[0057] S9. End.

[0058] Figure 1 The details of the difference frequency vocabulary, matching, authority identification, and speech noise reduction involved in the process are described below.

[0059] In the above process, the level priority matching must be based on the difference frequency special word library. The difference frequency special word library is automatically constructed according to the following principles of this patent:

[0060] The importance of special words in power dispatching is higher than that of common words, so its recognition rate should be guaranteed first, so a special word library should be established. Furthermore, the noisy environment affects the similarity of matching, so when the matching similarity is not much different, high-level special words should be matched first. In addition, the system should automatically build a library through a program. In order for the system to automatically build a library through a program and automatically distinguish special words from common words, its difference must be utilized. Special words, especially local special words, such as the special words mentioned above, such as "Yuegang Xiangshi Line", generally do not appear in ordinary news or articles, but appear in local power dispatching logs, work tickets, equipment records and local call text records. On the contrary, the general words mentioned above, such as "investment", will appear frequently in ordinary articles or online articles. In addition, the word "knife switch" may appear in local texts, power academic articles and news reports. Therefore, this patent proposes that the level of a word is determined by the difference between its two frequencies, that is, the higher the frequency of appearance in special materials, the higher the level of the word, and the higher the frequency of appearance in general materials, the lower the level of the word.

[0061] The automatic construction method of the difference frequency special word library is as follows:

[0062] S1. Collect vocabulary of specialized data, including local power dispatch logs, work tickets, equipment records and call texts, and search for power professional academic articles on the Internet;

[0063] S2. Clean and segment the vocabulary of the special data collected in the previous step, and obtain a vocabulary list, then count the word frequency of the vocabulary list to obtain a special word frequency dictionary. The word frequency = the number of times the word is repeated × the length of the word / the total number of words in all the data; the local data contains not only special vocabulary (i.e. local special vocabulary and professional terms) but also general vocabulary, such as the above example "put into use Yuetang Station Yuegang Xiangshi Line 35 grounding knife switch and 36 grounding knife switch", so the special word frequency dictionary contains both special vocabulary and general vocabulary;

[0064] S3. Statistical analysis of the text vocabulary of the People's Daily corpus and the news of the three major websites Sina, Sohu, and Netease was performed to obtain a general word frequency dictionary. The word frequency calculation method is the same as above. The general data mainly contains general words, and may also contain some professional terms, but the frequency of occurrence is low. For example, there may be "grounding knife switch" in news reports on electricity. Generally, local special words, such as "Yuegang Xiangshi Line", rarely appear.

[0065] S4. Perform a difference frequency operation on each word in the special word frequency dictionary to obtain a difference frequency special word library:

[0066] Difference frequency The special word frequency of a word - k × its general word frequency, where k is a fixed coefficient and can be obtained through statistics; the special word frequency and general word frequency are obtained by looking up the special word frequency dictionary and the general word frequency dictionary respectively. The word frequency that cannot be found in the dictionary is counted as 0; Example 1, to calculate the difference frequency value of "Yuegang Xiangshi Line", first look up its special word frequency in the special word frequency dictionary, and then if the word cannot be found in the general word frequency dictionary, its general word frequency is 0; Example 2, to calculate the difference frequency value of "earthing knife switch", first look up its special word frequency in the special word frequency dictionary, then look up its general word frequency in the general word frequency dictionary, and finally calculate its difference frequency value;

[0067] S5. The top 25% of the words with the difference frequency value are stored in the first-level sub-bank, the words with the difference frequency value of 26% to 50% are stored in the second-level sub-bank, the words with the difference frequency value of 51% to 75% are stored in the third-level sub-bank, and the rest of the words with the difference frequency value greater than 0 are stored in the fourth level. The words with the difference frequency value less than or equal to 0 are discarded. In the same level of sub-bank, the words with the higher difference frequency value are ranked first;

[0068] S6. Associate each word in the frequency difference special word library with its corresponding pinyin, and finally form a frequency difference special word library for power dispatching. The frequency difference special word library is composed of the first, second, third and fourth level sub-libraries.

[0069] exist Figure 1 In the process, the matching of pinyin, vocabulary and text can be carried out according to a known method. The present invention provides a preferred matching scheme comprising the following steps:

[0070] Calculate the phoneme edit distance: refers to the minimum number of phoneme edit operations required to convert one pinyin string into another. The phoneme refers to the initial consonant or final vowel of pinyin. The permitted editing operations include: inserting an initial consonant / final vowel, deleting an initial consonant / final vowel, replacing one initial consonant / final vowel with another. The replacement between ambiguous sounds is only counted as 0.5 times. For example, assuming that the speaker's Mandarin is not standard, Yuetang Station "yue4 tang2 zhan4" is said as "yue4 tan2 zhan4", then the correct pinyin can be obtained by replacing a final vowel ang. Here an and ang are mutually ambiguous sounds, so the phoneme edit distance is 0.5.

[0071] If the matching word is a special word, then when the phoneme edit distance is less than a given threshold, the phoneme edit distance and a match success signal are output, otherwise a match failure signal is given;

[0072] If the match is a common vocabulary, the phoneme edit distance is output.

[0073] The tones of Pinyin are not considered here, because there are many Chinese dialects and the pronunciations vary greatly from place to place. Many people find it difficult to distinguish tones, and changes in intonation and tone also affect the tones.

[0074] exist Figure 1 The process involves permission identification. For details, see Figure 2 As shown, the process is as follows:

[0075] 1) Get the wake-up voice. The voice is acquired through the microphone device and stored on the hard disk in the format of a .wav file. The computer reads the voice by calling the wave module in Python.

[0076] 2) Extract the Fbank energy features of each frame of the speech obtained in step 1). In this part, the python_speech_features module of Python is used to obtain the Fbank energy features of the speech.

[0077] 3) Input the Fbank energy features into a pre-trained DNN model and obtain a high-level feature expression based on deep learning. In this section, the DNN model architecture includes four hidden layers, which is implemented using the keras module of python.

[0078] 4) Perform cosine similarity matching between the speech feature expression obtained in step 3) and the speech feature of the user bound to the user login account; if the cosine similarity is lower than or equal to the threshold, it is judged that the user does not match the user bound to the login account and has no permission to use; if it is higher than the threshold, it is judged that the user has permission to use.

[0079] exist Figure 1 The process involves speech noise reduction. For details, see Figure 3 As shown, the process is as follows:

[0080] 1) Get the voice read by the user.

[0081] 2) Use the short-time Fourier transform (STFT) to extract the amplitude spectrum and phase spectrum of the speech obtained in step 1). In this step, the speech is divided into several small segments according to a 1-second window. Then, the short-time Fourier transform (STFT) provided by the librosa module of Python is used to obtain the amplitude spectrum and phase spectrum of each small segment.

[0082] 3) Input the amplitude spectrum obtained in step 2) into a pre-trained Unet model to obtain the denoised amplitude spectrum based on deep learning. Each small segment of the amplitude spectrum will be converted into a denoised amplitude spectrum through the Unet model. The Unet model is implemented using the keras module of python.

[0083] 4) The denoised amplitude spectrum obtained in step 3) and the phase spectrum obtained in step 2) are converted into denoised speech through inverse short-time Fourier transform (ISTFT). Through this step, the denoised speech corresponding to each short segment of speech is obtained, and then the denoised speech corresponding to the speech in step 1) is obtained through speech combination. The specific function of this step is implemented by the librosa module of python.

[0084] When the dispatcher and the on-site maintenance personnel have a voice conversation, there is often a lot of construction background noise, which seriously reduces the accuracy of speech recognition. In a noisy environment, some words may not be heard clearly. People can often guess some words that are not heard clearly from the context of the conversation, but Figure 1 The process only considers the recognition of a single sentence and cannot utilize the coherent topic semantics in the conversation context, which is also the weakness of the current speech recognition algorithm. A preferred solution is to add topic word matching before level priority matching to clarify the topic of the conversation and improve the recognition rate of the entire conversation. The specific steps are as follows:

[0085] Determine the subject words: count the words repeated in the previous text and their repetition times; if the repeated word is a frequency-delayed special word, add the word to the subject word queue, otherwise discard it.

[0086] Sorting of topic word queue: Assume that there are n dialogue sentences from the beginning of speech recognition to the current sentence to be recognized, and the current sentence to be recognized is numbered as sentence n+1, then the topic value of a repeated word is:

[0087] ;

[0088] In the formula, i and j are the repetitions of the word in the i-th and j-th sentences, and the ellipsis indicates other repeated sentences, i, j < n; G is the level of the difference frequency vocabulary sub-library to which the word belongs, and its value is an integer from 1 to 4. Calculate the topic values ​​of all topic words in the first n sentences, and then sort them from large to small according to the topic value.

[0089] Keyword matching: Match the pinyin string A obtained by speech recognition with the keyword queue, starting from the first keyword in the queue. If the match is successful, part of the pinyin of the pinyin string will be converted into text. If the match is unsuccessful, consider the next keyword until the last keyword in the queue.

[0090] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A speech recognition method for the field of power dispatching, characterized in that: include: Level priority matching: after converting speech to pinyin, a pinyin string consisting of letters and tones is obtained, which is named A. In the process of converting A into text, it is preferentially matched with the pinyin of the first-level sub-library vocabulary of the difference frequency special vocabulary. If the match is successful, part of the pinyin of A is converted into text. If the match is unsuccessful, the next level is considered, until the last level sub-library. The matching is to find the similarity between a part of the pinyin of A and the correct pinyin of a Chinese phrase or character, which is also referred to as the matching of pinyin and text or vocabulary. The difference frequency special vocabulary is composed of hierarchical special vocabulary and its pinyin. The level of the vocabulary is determined by the difference between its two frequencies, that is, the higher the frequency of occurrence in special materials, the higher the level of the vocabulary, and the higher the frequency of occurrence in general materials, the lower the level of the vocabulary. The vocabulary refers to Chinese phrases, and a Chinese phrase includes at least 2 Chinese characters. The special vocabulary includes local special vocabulary and professional terms. The local special vocabulary refers to vocabulary used only in the local machine, local area network, a specific region, group or department. Special vocabulary of the same level constitutes a sub-library. The highest level sub-library is the first level sub-library, and the following are from the second level to the lowest level sub-library. Frequency priority matching: After the level priority matching is completed, the remaining pinyin of A is matched with the pinyin of common words. Non-special words with high frequency appearing in common data are matched first, and finally the remaining pinyin is matched with the pinyin of a single Chinese character; The construction of the difference frequency special word library includes the following steps: S1. Collect vocabulary of specialized data, including local power dispatch logs, work tickets, equipment records and call texts, and search for power professional academic articles on the Internet; S2. Clean and segment the vocabulary of the special data collected in the previous step to obtain a vocabulary list, and then count the word frequency of the vocabulary list to obtain a special word frequency dictionary; wherein, word frequency = the number of times the word is repeated × the length of the word / the total number of words in all the data, and the special word frequency dictionary contains all the words and their word frequencies that appear in the special data; S3, collect vocabulary from general data, including the People's Daily corpus and news from three major websites, Sina, Sohu, and NetEase, and perform vocabulary frequency statistics to obtain a general word frequency dictionary, which includes all the vocabulary and word frequencies that appear in the general data; S4, perform a difference frequency operation on each word in the special word frequency dictionary to obtain a difference frequency special word library, wherein the difference frequency operation is: Difference frequency value The special word frequency of a word - k × its general word frequency, where k is a fixed coefficient. The special word frequency and general word frequency are obtained from the special word frequency dictionary and the general word frequency dictionary respectively. The word frequency that cannot be found in the dictionary is counted as 0; S5. The top 25% of the words with the difference frequency value are stored in the first-level sub-bank, the words between 26% and 50% are stored in the second-level sub-bank, the words between 51% and 75% are stored in the third-level sub-bank, and the rest of the words with a difference frequency value greater than 0 are stored in the fourth level. Words with a difference frequency value less than or equal to 0 are discarded; S6. Associate each word in the frequency difference special word library with its corresponding pinyin, and finally form a frequency difference special word library for power dispatching; the frequency difference special word library is composed of first, second, third and fourth level sub-libraries, and the higher the frequency difference value, the higher the word is in the queue of the sub-library.

2. A speech recognition method for the electric power dispatching field according to claim 1, characterized in that: The steps of obtaining a pinyin string composed of letters and tones include: S1. Authority identification: Perform voiceprint recognition on the wake-up voice input by the user. If the user has the authority to use, proceed to step S2, otherwise, exit; S2, speech noise reduction: perform speech noise reduction on user speech to reduce the interference of noise on recognition; S3, voice to pinyin conversion: perform voice analysis and recognition on the voice after noise reduction in step S2, and obtain the whole sentence pinyin string corresponding to the voice.

3. A speech recognition method for the field of power dispatching according to claim 2, characterized in that: The authority identification comprises the following steps: S11, obtaining a wake-up voice; S12, extracting the Fbank energy features of each frame in the speech obtained in step S11; S13, input the Fbank energy features into a pre-trained DNN model to obtain a high-level feature expression based on deep learning; S14. Perform cosine similarity matching between the high-level feature expression obtained in step S13 and the user voice feature bound to the user login account; if the cosine similarity is lower than or equal to the set value, it is determined that the user does not match the user bound to the login account and has no use permission; if it is higher than the set value, it is determined that the user has use permission.

4. A speech recognition method for the electric power dispatching field as claimed in claim 1, characterized in that: The level priority matching step includes: Reverse word selection: select the pinyin of the word with the highest difference frequency value from the words that have not been matched in the first-level sub-library, and set its name as B. If all the words in the first-level sub-library have been matched, then move on to the next-level sub-library; Convert pinyin at any position: Search for a substring C similar to B in A. If B matches C successfully, convert C into the corresponding Chinese phrase. If there are multiple substrings similar to B in A, repeat the above operation. The substring C can be located at any position in A.

5. A speech recognition method for the electric power dispatching field as claimed in claim 1, characterized in that: Also includes: When the recognized speech is a conversation, the subject word determination, subject word queue sorting and subject word matching are performed before the level priority matching, as follows: Determine the subject word: count the words repeated in the previous text and their repetition times; if the repeated word is a frequency-delayed special word, add the word to the subject word queue, otherwise discard it; Sorting of topic word queue: Assume that n conversation sentences have been recognized from the start of this speech recognition to the current sentence to be recognized, and the current sentence to be recognized is numbered as sentence n+1, then the topic value of a repeated word is: ； In the formula, i and j are the words repeated in the i-th and j-th sentences, and the ellipsis indicates other repeated sentences; i, j < n; G is the level of the sub-library of the difference frequency special word library to which the word belongs, and its value is an integer from 1 to 4; calculate the topic values ​​of all the topic words in the first n sentences, and then arrange them from large to small according to the topic values ​​to obtain the topic word queue; Subject word matching: Match A with the subject word queue, starting from the first subject word in the queue. If the match is successful, part of the pinyin of A will be converted into text. If the match is unsuccessful, consider the next subject word until the last subject word in the queue.

6. A speech recognition method for the electric power dispatching field as claimed in claim 1, 4 or 5, characterized in that: The matching comprises the following steps: Calculate the phoneme edit distance: refers to the minimum number of phoneme edit operations required to convert one pinyin string into another. The phoneme refers to the initial consonant or final of the pinyin. The permitted edit operations include: inserting an initial consonant or final, deleting an initial consonant or final, replacing one initial consonant or final with another. The replacement between ambiguous sounds is only counted as 0.5 times. The above operations do not include tones. If the matching word is a special word, then when the phoneme edit distance is less than a given threshold, the phoneme edit distance and a match success signal are output, otherwise a match failure signal is given; If the match is a common vocabulary, the phoneme edit distance is output.

Images