A speech-to-text error correction method, device, electronic device, and storage medium
By replacing the original text and recognizing new words and adjusting new words in the speech-to-text, the problem of errors in the recognition of new words in offline systems is solved, the accuracy and user experience of speech-to-text are improved, and self-learning and self-learning are realized. optimization.
Patent Information
- Application Number
- CN202111181390.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-11
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-10-11
AI Technical Summary
The offline conference recording system cannot recognize new words, resulting in errors in speech to text, and lack of self-learning and self-optimization capabilities, which increases the work burden of recorders.
By replacing the original text, identifying new words and adjusting according to adjustment instructions, target text is generated, wrong words are reduced, and the accuracy of speech to text is improved.
It improves the accuracy of voice to text, reduces manual operations by recorders, realizes self-learning and self-optimization functions, and improves user experience and work efficiency.
Smart Images

Figure CN114822527B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of speech-to-text conversion, and in particular to a speech-to-text conversion error correction method, device, electronic device and storage medium. Background Art
[0002] Automated devices or tools for meeting records (i.e., meeting record systems) are suitable for scenarios such as multi-person meetings. In a multi-person meeting, participants may speak at different time periods. The meeting record system can collect voice through a microphone and transcribe the collected voice into text-based meeting speech records. High-quality transcription results can significantly improve the user experience of the relevant meeting record system. Different types of meetings have different requirements for meeting record systems. Some meetings can only use offline versions of meeting record systems for information security and confidentiality considerations. At present, the common offline meeting record systems on the market have the following problems:
[0003] (1) In today's rapidly developing world, new words may appear in meetings. If you want to identify these new words as accurately as possible, you need to update the vocabulary frequently. For ordinary conference systems, updating the vocabulary online is a convenient and efficient solution. However, offline conference recording systems cannot connect to the Internet, and even manually updating the vocabulary requires a complex process, which greatly increases the difficulty of updating the vocabulary. In this case, speech-to-text may not give correct results when encountering new words, but will give results containing errors or transcribe them as incorrect words. For such transcription errors, the current common conference records mainly rely on manual modification. Before modifying them in combination with the context and related audio, the recorder needs to find the location of the error by himself. This work that requires manual proofreading of the entire text is a heavy burden for recorders of large meetings.
[0004] (2) After the recorder corrects the location of the transcription error, if he wants the system to automatically correct the wrong word in the future, he needs to manually add it as a template. In contrast, if the meeting record system can automatically predict the words that can be added as templates based on the recorder's modification behavior, it can reduce the recorder's manual operations and significantly improve work efficiency. However, the current offline meeting record system lacks the ability of self-learning and self-optimization and cannot achieve such functions.
[0005] Therefore, how to solve the problem that speech-to-text cannot recognize new words in offline state, resulting in speech-to-text errors, has become an urgent problem to be solved. Summary of the invention
[0006] In order to solve the above technical problem or at least partially solve the above technical problem, the present invention provides a speech-to-text error correction method, device, electronic device and storage medium.
[0007] In a first aspect, the present invention provides a speech-to-text error correction method, the speech-to-text error correction method comprising: replacing an original text to obtain a modified text, the original text being a text converted according to a target speech; identifying new words in the modified text and collecting the identified new words; and adjusting the new words in the modified text according to an adjustment instruction to obtain a target text.
[0008] In a second aspect, the present invention provides a speech-to-text error correction device, the speech-to-text error correction device comprising: a replacement module, used to replace an original text to obtain a modified text, wherein the original text is a text converted according to a target speech; a recognition module, used to identify new words in the modified text and collect the identified new words; an adjustment module, used to adjust the new words in the modified text according to an adjustment instruction to obtain a target text.
[0009] In a third aspect, this embodiment further provides an electronic device, characterized in that it includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the steps of the speech-to-text error correction method as described above when executing the program stored in the memory.
[0010] In a fourth aspect, the present invention further provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the speech-to-text error correction method as described above.
[0011] The above technical solution provided by the embodiment of the present invention has the following advantages compared with the prior art:
[0012] The speech-to-text error correction method provided by the embodiment of the present invention comprises: replacing an original text to obtain a modified text, wherein the original text is a text converted according to a target speech; performing new word recognition on the modified text and collecting the recognized new words; adjusting the new words in the modified text according to an adjustment instruction to obtain a target text, by processing the original text and then performing new word recognition on the obtained modified text, by recognizing the new words converted from the target speech and adjusting them, in this process, adjusting the wrong words in the new words, thereby reducing the wrong words in the modified text, thereby improving the accuracy of speech-to-text conversion, and thereby improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0015] Figure 1 A basic flow chart of a speech-to-text error correction method provided by an embodiment of the present invention;
[0016] Figure 2 A basic structural diagram of a speech-to-text error correction device provided by an embodiment of the present invention;
[0017] Figure 3 A basic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0019] In order to solve the problem that speech-to-text cannot recognize new words in offline state, resulting in speech-to-text errors, this embodiment provides a speech-to-text error correction method, such as Figure 1 As shown, the speech-to-text error correction method includes:
[0020] S101, replacing an original text to obtain a modified text, wherein the original text is a text converted according to a target voice;
[0021] It should be understood that the original text is a text converted from the target voice, wherein the target voice includes but is not limited to: conference voice, chat voice, call voice, etc. At the same time, the speech-to-text error correction method provided in this embodiment can be implemented in various forms, for example, it can be implemented through a terminal, and the terminal described in the present invention may include mobile terminals such as mobile phones, tablet computers, laptop computers, PDAs, portable media players (PMPs), navigation devices, etc., as well as fixed terminals such as digital TVs and desktop computers. It should be understood that voice recorders, conference recording devices, etc. are also included in the scope of the above-mentioned terminals.
[0022] S102, identifying new words in the modified text, and collecting the identified new words;
[0023] It needs to be understood that after obtaining the modified text, it is necessary to identify the new words in the modified text. To extract the new words in the modified text, it needs to be understood that new word discovery includes a variety of methods, which can be used alone or in combination to identify the new words in the modified text. After collecting the identified new words, the new words in the modified text can be adjusted according to the adjustment instructions.
[0024] S103: adjusting the new words in the modified text according to the adjustment instruction to obtain a target text.
[0025] It is important to understand that the new words are adjusted, that is, the errors in the speech-to-text are corrected, to obtain the target text.
[0026] In some examples of this embodiment, replacing the original text to obtain the modified text includes: preprocessing the original text to remove redundant content in the original text; inputting the preprocessed original text as input text into the replacement model to obtain the modified text, the replacement model is used to replace the words in the input text that meet the replacement requirements using a preset replacement word library, and then outputting the replaced text as the modified text. It should be understood that the original text is converted from speech, which may include redundant content. Therefore, the redundant content needs to be removed, and then the existing replacement word library is used to automatically find the words that need to be replaced in the original text, and the original text is preliminarily corrected to obtain the modified text, wherein the words that need to be replaced are also words that can be corrected according to the existing word library; for example, taking conference speech as an example, the original text converted from the conference speech may contain redundant punctuation marks and debugging text that is irrelevant to the conference content, etc. First, the original text is preprocessed to remove redundant punctuation marks and debugging text that is irrelevant to the conference content, and then the existing replacement word library is used to automatically find the words that need to be replaced in the original text, and the original text is preliminarily corrected to obtain the modified text.
[0027] In some examples of the present embodiment, new word recognition is performed on the modified text, including: determining whether there are words that conform to the rule template in the modified text; when there are words that conform to the rule template in the modified text, the words that conform to the rule template are used as new words. Among them, the rule template is a preset template; it should be understood that different types of text have different text features, which causes the features of new words in the text to be different. Therefore, the preset rule template can intuitively fit different types of text, that is, different types of text correspond to different preset rule templates, thereby improving the work efficiency of speech-to-text conversion and the performance of identifying new words.
[0028] Continuing with the previous example, taking the text of the conference voice as an example, we can analyze it to obtain the rule template of the conference text. The specific steps are as follows:
[0029] Step 1: Analyze the text converted from speech and summarize appropriate rules. Phrases that summarize the nature of the meeting often appear in meetings, such as "three xx", "four yy", etc. When new words appear, such phrases appear more frequently. From the perspective of phrase structure, such phrases generally start with a quantifier, and the words after the quantifier will appear multiple times in the nearby context. For other types of meetings, their new words may have different forms of expression. Therefore, the preset rule template is not limited to phrase structure, but can also include word group combinations, grammatical structures, etc.
[0030] Step 2: Express the found rules with regular expressions to obtain a rule template.
[0031] Through the above steps 1 and 2, the preset rule template is obtained, and then the regular expression is used to check whether the modified text contains parts that meet the corresponding rules. The matched words are used as new words, and the new words in the modified text are further identified.
[0032] In some examples of this embodiment, when there are words that conform to the rule template in the modified text, the words that conform to the rule template are taken as new words, and it also includes: when there are words that conform to the rule template in the modified text, the text that conforms to the rule template is compared with the words in the standard words to determine whether the words that conform to the rule template appear in the standard vocabulary; when the words that conform to the rule template do not appear in the standard vocabulary, the words that conform to the rule template are taken as new words. It should be understood that the words of the rule template may belong to existing words. Therefore, when determining whether there are words that conform to the rule template in the modified text, it is also necessary to determine whether they belong to existing vocabulary. Specifically, when the words that conform to the rule template appear in the standard vocabulary, it means that the word is not a new word. Therefore, only when the words of the modified text successfully match the rule template and the matched words do not appear in the existing standard vocabulary, will the word be determined to be a new word.
[0033] In some examples of this embodiment, the new word recognition of the modified text includes: performing word frequency statistics on each word in the modified text to determine the word frequency score of each word; comparing the words whose word frequency scores reach a preset score with the words in the standard word library to determine whether the words whose word frequency scores reach a preset score appear in the standard word library; and treating the words whose word frequency scores reach a preset score and do not appear in the standard word library as new words. That is, new words can be screened by calculating the information entropy and point-to-point mutual information of the words in the modified text. Information entropy represents the richness of the adjacent characters of a word, and point-to-point mutual information can be used to measure the degree of cohesion within the word.
[0034] In some examples of the present embodiment, word frequency statistics are performed on each word in the modified text to determine the word frequency score of each word, including: dividing the modified text into multiple words; calculating the left and right information entropy scores and mutual information scores of each word; and performing comprehensive calculations on the left and right information entropy scores and mutual information scores of each word to obtain the word frequency score of each word.
[0035] Continuing from the above example, in some examples, the specific steps of performing word frequency statistics on each word in the modified text and determining the word frequency score of each word are as follows:
[0036] Step 1: Split the modified text word by word and then concatenate them to generate multiple words. When generating words, you need to set the maximum length of the words. For different types of text, the new words may also be different, so the appropriate length of the words segmented from the modified text is also different. For example, the new words of some meetings are usually longer than those of other types of meetings.
[0037] Step 2: Calculate the left and right information entropy scores for the words. The calculation formulas for the left and right information entropy Entropy(wl) and Entropy(wr) of word w are as follows:
[0038]
[0039]
[0040] Among them, W NL and W NR are the sets of adjacent characters on the left and right of w. A word should be able to be matched with multiple different contexts, so a qualified word should have large word information entropy on both the left and right sides. Therefore, it is necessary to optimize this formula and merge the left and right information entropy into one formula. The calculation formula after optimization and merging is:
[0041]
[0042] Step 3: Calculate the mutual information score of the word. The calculation formula of the point-to-point mutual information PMI(w) of word w is:
[0043]
[0044] Where c1, c2, ..., c n are the characters that make up word w, and P represents the probability of a character or word appearing in the text.
[0045] From this formula, we can see that when the word becomes longer, the inter-point mutual information score of the new word tends to increase due to the small probability of occurrence of the newly added characters P(c). In order to suppress this trend, it is necessary to modify the formula appropriately. After the modification, the calculation formula of the inter-point mutual information APMI(w) of word w is:
[0046] APMI(w)=(n+1)log(P(w))-log(P(c1)·P(c2)·...·P(c n ))
[0047] Step 4: Combine the above two scores using different weights:
[0048] Score(w)=p·Entropy(w)+q·APMI(w)
[0049] Among them, p and q are coefficients. By adjusting the coefficients, the weights of information entropy and mutual information between points can be changed. Use this formula to calculate the total score of each word and remove the words with lower scores. Specifically, the total scores of each word can be sorted, and the words with the lowest ranking of 30%-70% can be removed, and then the words with lower scores can be removed; or a score can be preset, and all words that do not meet the preset score can be removed, and then the words with lower scores can be removed; and then the remaining words are used as new words.
[0050] Continuing from the above example, it is necessary to understand that the remaining words may also be existing words. Therefore, it is necessary to determine whether the remaining words belong to existing vocabulary. Specifically, when the remaining words appear in the standard vocabulary, it means that the word is not a new word. Therefore, only when the remaining words do not appear in the existing standard vocabulary will the word be determined to be a new word.
[0051] In some examples of this embodiment, adjusting the new words in the modified text according to the adjustment instruction to obtain the target text includes: determining the adjustment method corresponding to the adjustment instruction, the adjustment method includes: at least one of adding, deleting, and replacing; adjusting the new words in the modified text according to the adjustment method, and using the modified text formed by the adjustment as the target text. Specifically, the adjustment instruction can be an adjustment instruction input by the user, and the adjustment instruction can also be an adjustment instruction automatically generated by the system according to the user's previous behavior. For example, when the new words in the modified document are collected through the above-mentioned recognition step, the collected new words are highlighted to the user (for example: the meeting recorder), so that the meeting recorder no longer needs to check the modified text word by word, thereby reducing the workload. The meeting recorder can first check the transcription of the new words and make modifications to possible errors, and the modification methods include adding, deleting and replacing. It should be understood that for other parts of the modified text, the meeting recorder can also make modifications in the modified text. When the meeting recorder completes the modification, the system saves the modified text as the target text and saves the modification behavior of the meeting recorder at the same time.
[0052] In some examples of this embodiment, the modification of the new words in the previous step is an important reference. The modified new words can be added to the replacement word library, wherein the specific determination of which new words are added to the replacement word library can be achieved by statistical analysis of the modification records of the modified text, and the specific method is as follows:
[0053] Step 1: Calculate the probability of each new word being changed and the probability of its corresponding target word by changing the modified text. When a new word w in the modified text is replaced, its replacement probability P(cw) is:
[0054] P(cw)=F(w′) / F(w)
[0055] Among them, F(w) is the frequency of w appearing, and F(w′) is the frequency of w being replaced by other words.
[0056] The word used to replace w belongs to the set K. For each word k in K, the target word probability P(kw) is:
[0057] P(kw)=F(kw) / F(w′)
[0058] Among them, F(kw) is the frequency of k replacing w.
[0059] Continuing with the above example, for example: if there are 10 "three xxy"s and 10 "three zy"s in the new words of the modified text, and the adjustment instruction changes 6 "three xxy"s and 8 "three zy"s into "three xy", and replaces 1 "three xxy" into "three axy", and the other new words are not changed, then the replacement probability of "three xxy" is (6+1) / 10=0.7, the replacement probability of "three zy" is 8 / 10=0.8, the target word probability of "three xy" for "three xxy" is 6 / (6+1)=0.857, the target word probability of "three axy" for "three xxy" is 1 / (6+1)=0.143, and the target word probability of "three xy" for "three zy" is 8 / 8=1.
[0060] Step 2: For each new word w and each replaced word k, use the above two probabilities to calculate the score:
[0061] Score(w,k)=a·(P(cw)-c)+b·(P(kw)-d)
[0062] Among them, a and b are weight coefficients, and c and d are the minimum standards for replacement probability and target word probability.
[0063] Step 3. Set two score lines p and q according to the above scores, p>q. When P(cw)>c and P(kw)>d, if Score(w,k)>=p, the system will automatically add the replacement word pair (w,k) directly to the replacement word library; if p>Score(w,k)>=q, the system will ask the meeting recorder whether to add the replacement word pair (w,k) to the word library.
[0064] Step 4: The expanded replacement word library can be used to directly replace the new words appearing in the original text to obtain a modified text.
[0065] The speech-to-text error correction method provided in the present embodiment obtains a modified text by replacing the original text, wherein the original text is a text converted from a target speech; new words are recognized on the modified text, and the recognized new words are collected; the new words in the modified text are adjusted according to the adjustment instruction to obtain the target text, by processing the original text, and then new words are recognized on the obtained modified text, by recognizing the new words converted from the target speech and adjusting them. In this process, the wrong words in the new words are adjusted, thereby reducing the wrong words in the modified text, thereby improving the accuracy of speech-to-text, and thereby improving the user experience. At the same time, in some examples, the adjusted and corrected words are put into a replacement word library. When speech-to-text is performed next time, the replacement word library can automatically replace the erroneous words, thereby improving the accuracy of speech-to-text and the efficiency of replacing the word library construction, thereby improving the error correction efficiency of speech-to-text.
[0066] like Figure 2 As shown, this embodiment also provides a speech-to-text error correction device, which includes but is not limited to:
[0067] A replacement module 1 is used to replace the original text to obtain a modified text, wherein the original text is a text converted according to the target voice;
[0068] Recognition module 2, used for identifying new words in the modified text and collecting the identified new words;
[0069] The adjustment module 3 is used to adjust the new words in the modified text according to the adjustment instruction to obtain the target text.
[0070] like Figure 3 As shown, an embodiment of the present application provides an electronic device, including a processor 111, a communication interface 112, a memory 113 and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114.
[0071] A memory 113, for storing computer programs;
[0072] In one embodiment of the present application, the processor 111 is used to implement the steps of the speech-to-text error correction method provided in any of the aforementioned method embodiments when executing the program stored in the memory 113.
[0073] An embodiment of the present application also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the speech-to-text error correction method provided in any of the aforementioned method embodiments are implemented.
[0074] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0075] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.
Claims
1. A speech-to-text error correction method, It is characterized in that The speech-to-text error correction method comprises: Replacing the original text to obtain a modified text, wherein the original text is a text converted according to the target voice; Performing new word recognition on the modified text and collecting the recognized new words; Adjusting the new words in the modified text according to the adjustment instruction to obtain the target text; The modified text obtained by replacing the original text includes: Preprocessing the original text to remove redundant content in the original text; The preprocessed original text is input as input text into the replacement model to obtain a modified text, wherein the replacement model is used to replace words in the input text that meet the replacement requirements using a preset replacement word library, and then output the replaced text as the modified text; Performing new word recognition on the modified text includes: Performing word frequency statistics on each word in the modified text to determine a word frequency score for each word; Compare the words whose frequency scores reach the preset score with the words in the standard word library to determine whether the words whose frequency scores reach the preset score appear in the standard word library; The words whose frequency scores reach the preset scores and do not appear in the standard word library are regarded as new words; Performing word frequency statistics on each word in the modified text to determine the word frequency score of each word includes: dividing the modified text into a plurality of words; Calculate the left and right information entropy scores and mutual information scores of each word. The left and right information entropy is determined by the following formula: Among them, Entropy(wl) is the left information entropy, Entropy(wr) is the right information entropy, w NL and w NR are the sets of adjacent characters on the left and right of word w respectively; The mutual information score is determined by: APMI(w)=(n+1)log(P(w))-log(P(c 1 )·P(c 2 )·...·P(c n )) Among them, c 1 , c 2 , ..., c n are the characters that make up word w, and P represents the probability of a character or word appearing in the text; The left and right information entropy scores and mutual information scores of each word are comprehensively calculated to obtain the word frequency score of each word.
2. The speech-to-text error correction method according to claim 1, It is characterized in that Performing new word recognition on the modified text, including: Determining whether there are words that conform to the rule template in the modified text; When there are words that conform to the rule template in the modified text, the words that conform to the rule template are used as new words.
3. The speech-to-text error correction method according to claim 2, It is characterized in that When there are words that conform to the rule template in the modified text, the words that conform to the rule template are used as new words, and the method further includes: When there are words that conform to the rule template in the modified text, the text that conforms to the rule template is compared with words in the standard word library to determine whether the words that conform to the rule template appear in the standard word library; When the word that meets the rule template does not appear in the standard word library, the word that meets the rule template is used as a new word.
4. The speech-to-text error correction method according to any one of claims 1 to 3, It is characterized in that Adjusting the new words in the modified text according to the adjustment instruction to obtain the target text includes: Determine an adjustment method corresponding to the adjustment instruction, where the adjustment method includes: at least one of adding, deleting, and replacing; The new words in the modified text are adjusted according to the adjustment method, and the modified text formed by the adjustment is used as the target text.
5. A speech-to-text error correction device, It is characterized in that The speech-to-text error correction device comprises: A replacement module, used to replace the original text to obtain a modified text, wherein the original text is a text converted according to the target voice; A recognition module, used for identifying new words in the modified text and collecting the identified new words; An adjustment module, used for adjusting the new words in the modified text according to the adjustment instruction to obtain a target text; The modified text obtained by replacing the original text includes: Preprocessing the original text to remove redundant content in the original text; The preprocessed original text is input as input text into the replacement model to obtain a modified text, wherein the replacement model is used to replace words in the input text that meet the replacement requirements using a preset replacement word library, and then output the replaced text as the modified text; Performing new word recognition on the modified text includes: Performing word frequency statistics on each word in the modified text to determine a word frequency score for each word; Compare the words whose frequency scores reach the preset score with the words in the standard word library to determine whether the words whose frequency scores reach the preset score appear in the standard word library; The words whose frequency scores reach the preset scores and do not appear in the standard word library are regarded as new words; Performing word frequency statistics on each word in the modified text to determine the word frequency score of each word includes: dividing the modified text into a plurality of words; Calculate the left and right information entropy scores and mutual information scores of each word. The left and right information entropy is determined by the following formula: Among them, Entropy(wl) is the left information entropy, Entropy(wr) is the right information entropy, w NL and w NR are the sets of adjacent characters on the left and right of word w respectively; The mutual information score is determined by: APMI(w)=(n+1)log(P(w))-log(P(c 1 )·P(c 2 )·...·P(c n )) Among them, c 1 ,c 2 , ..., c n are the characters that make up word w, and P represents the probability of a character or word appearing in the text; The left and right information entropy scores and mutual information scores of each word are comprehensively calculated to obtain the word frequency score of each word.
6. An electronic device, It is characterized in that It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, for implementing the steps of the speech-to-text error correction method as described in any one of claims 1 to 4 when executing a program stored in a memory.
7. A storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the speech-to-text error correction method according to any one of claims 1 to 4 are implemented.
