Speech recognition post-processing method and device, and computer-readable storage medium

Abstract

The present invention discloses a post-processing method for speech recognition and its device, and a computer-readable storage medium. The method comprises: performing phoneme analysis on a target text to obtain a plurality of first words with the same phonemes in the target text; determining some of the first words whose part of speech in the target text is a noun; analyzing the subtext where some of the first words are located to obtain an analysis result; when the analysis result indicates that the first words in the subtext are different, determining a second word in the subtext to replace the first word; and replacing the first word in the subtext with the second word. The present invention solves the technical problem of insufficient recognition of words with the same phonemes but different contents in the speech recognition process in the related art.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing, and in particular to a speech recognition post-processing method and device thereof, and a computer-readable storage medium. Background Art

[0002] Existing speech recognition technology is generally divided into two stages: first, audio data is converted into phonemes through an acoustic model; second, the phoneme data is converted into matching text recognition results through a language model. In the second stage of the above speech recognition technology, in order to ensure the real-time rate, it is difficult to decode phonemes into text in the context of a large amount of text. As a result, the speech recognition results can only achieve the local optimal recognition results, and cannot achieve the global optimal recognition results.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present invention provide a post-processing method and device for speech recognition, and a computer-readable storage medium, so as to at least solve the technical problem of insufficient recognition of words with the same phoneme but different contents in the speech recognition process in the related art.

[0005] According to one aspect of an embodiment of the present invention, a post-processing method for speech recognition is provided, comprising: performing phoneme analysis on a target text to obtain a plurality of first words having the same phonemes in the target text; determining some of the first words whose part of speech in the target text is noun; analyzing a sub-text where the some of the first words are located to obtain an analysis result; when the analysis result indicates that the first words in the sub-text are different, determining a second word in the sub-text to replace the first word; and replacing the first word in the sub-text with the second word.

[0006] Optionally, before performing phoneme analysis on the target text, the method further includes: converting the original sound signal into phoneme information through an acoustic model; and decoding the phoneme information using a language model to obtain the target text.

[0007] Optionally, determining the part of the first words whose part of speech is noun in the target text includes: locating a sub-text in the target text where the multiple first words are located, wherein the sub-text includes at least one of the following: a phrase, a sentence, and a paragraph; performing part-of-speech analysis on the sub-text to obtain the part of the first words whose part of speech is noun in the sub-text.

[0008] Optionally, after obtaining the part of the first words whose part of speech is noun in the sub-text, the method further includes: adding a tag to the sub-text.

[0009] Optionally, the method further includes: when the analysis result indicates that the first words in the sub-texts are all the same, releasing the mark of the sub-text.

[0010] Optionally, determining the second word to replace the first word in the sub-text includes: processing the part of the first word based on a predetermined algorithm to obtain the second word, wherein the predetermined algorithm includes at least one of the following: a probability analysis algorithm and a voting selection algorithm.

[0011] According to another aspect of an embodiment of the present invention, a post-processing device for speech recognition is also provided, including: a first analysis module, used to perform phoneme analysis on a target text to obtain multiple first words with the same phonemes in the target text; a first determination module, used to determine some of the multiple first words whose part of speech in the target text is noun; a second analysis module, used to analyze the sub-text where the some of the first words are located to obtain an analysis result; a second determination module, used to determine a second word in the sub-text to replace the first word when the analysis result indicates that the first word in the sub-text is different; and a replacement module, used to replace the first word in the sub-text with the second word.

[0012] Optionally, the device further includes: a conversion module, used to convert the original sound signal into phoneme information through an acoustic model before performing phoneme analysis on the target text; and a decoding module, used to decode the phoneme information using a language model to obtain the target text.

[0013] Optionally, the first determination module includes: a positioning unit, used to locate the sub-text where the multiple first words are located in the target text, wherein the sub-text includes at least one of the following: a phrase, a sentence and a paragraph; an analysis unit, used to perform part-of-speech analysis on the sub-text to obtain the part of the first words in the sub-text whose part of speech is noun.

[0014] Optionally, the device further comprises: a marking module, configured to add a mark to the sub-text after acquiring the first part of the words whose part of speech is noun in the sub-text.

[0015] Optionally, the device further comprises: a releasing module, configured to release the mark of the sub-text when the analysis result indicates that the first words in the sub-text are all the same.

[0016] Optionally, the second determination module includes: a processing unit, configured to process the part of the first words based on a predetermined algorithm to obtain the second words, wherein the predetermined algorithm includes at least one of the following: a probability analysis algorithm and a voting selection algorithm.

[0017] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is also provided, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed by a processor, the device where the computer-readable storage medium is located is controlled to perform any of the above-mentioned post-processing methods for speech recognition.

[0018] According to another aspect of the embodiments of the present invention, a processor is provided, wherein the processor is used to run a computer program, wherein the computer program executes any one of the above-mentioned post-processing methods for speech recognition when running.

[0019] In an embodiment of the present invention, a target text is subjected to phoneme analysis to obtain multiple first words with the same phonemes in the target text; some of the first words whose part of speech in the target text is a noun are determined; the subtext where some of the first words are located is analyzed to obtain an analysis result; when the analysis result indicates that the first words in the subtext are different, a second word is determined in the subtext to replace the first word; and the first word in the subtext is replaced by the second word. The post-processing method for speech recognition provided by the embodiment of the present invention achieves the purpose of extracting words with the same pronunciation but different characters by utilizing the characteristics of the same phonemes and correcting the text by utilizing text analysis, thereby achieving the technical effect of improving the accuracy of speech recognition, and further solving the technical problem of insufficient recognition of words with the same phonemes but different contents in the speech recognition process in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0021] Figure 1 is a flowchart of a post-processing method for speech recognition according to an embodiment of the present invention;

[0022] Figure 2 is a schematic diagram of a post-processing device for speech recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] To enable those skilled in the art to better understand the solution of the present invention, the following will clearly and completely describe the technical solution in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0025] Embodiment 1

[0026] According to an embodiment of the present invention, a method embodiment of a post-processing method for speech recognition is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from that here.

[0027] Figure 1 is a flowchart of a post-processing method for speech recognition according to an embodiment of the present invention. As Figure 1 shown, the method includes the following steps:

[0028] Step S102, perform phoneme analysis on the target text to obtain multiple first words with the same phonemes in the target text.

[0029] Optionally, in the above step, analyze the text converted from speech, and extract the words with the same phonemes but different contents (i.e., the first words) therein. For example, the phonemes of "一一" and "依依" are both "yi1yi1", but they have different contents, that is, different font expressions. In addition, it should be noted that there can be multiple words with the same phonemes but different contents extracted.

[0030] Step S104, determine a part of the first words among the multiple first words whose part of speech in the target text is a noun.

[0031] Optionally, in the above steps, the parts of speech of the words with the same phonemes but different contents extracted above are analyzed, and those with the part of speech of nouns are extracted for secondary processing.

[0032] Step S106, analyzing the subtext where part of the first word is located to obtain an analysis result.

[0033] Step S108: when the analysis result indicates that the first words in the sub-texts are different, determine a second word in the sub-texts to replace the first word.

[0034] Optionally, in the above steps, when the analysis result indicates that multiple words with the same phonemes but different contents extracted in the sub-text are different, a predetermined algorithm is used to determine a word with the same phonemes as the first word that is the correct expression for correcting the first word.

[0035] Step S110: replacing the first word in the subtext with the second word.

[0036] As can be seen from the above, in the embodiment of the present invention, the target text can first be subjected to phoneme analysis to obtain multiple first words with the same phonemes in the target text; then, some of the first words whose part of speech in the target text is a noun can be determined; then, the subtext where some of the first words are located can be analyzed to obtain an analysis result; then, when the analysis result indicates that the first words in the subtext are different, a second word can be determined in the subtext to replace the first word; finally, the first word in the subtext can be replaced by the second word. The post-processing method for speech recognition provided by the embodiment of the present invention achieves the purpose of extracting words with the same pronunciation but different characters by utilizing the characteristics of the same phonemes and correcting the text by utilizing text analysis, thereby achieving the technical effect of improving the accuracy of speech recognition, and further solving the technical problem of insufficient recognition of words with the same phonemes but different contents in the speech recognition process in the related art.

[0037] As an optional embodiment, before performing phoneme analysis on the target text, the post-processing method of speech recognition also includes: converting the original sound signal into phoneme information through an acoustic model; and decoding the phoneme information using a language model to obtain the target text.

[0038] In the above optional embodiment, the original sound to be recognized (i.e., the original sound signal) is converted into pure phoneme information (i.e., phoneme information) through a set acoustic model, and then the phoneme information is decoded using a language model, and then converted into target text information, so that the target text can be corrected later.

[0039] As an optional embodiment, determining some first words whose part of speech is noun in a target text includes: locating a sub-text where the multiple first words are located in the target text, wherein the sub-text includes at least one of the following: a phrase, a sentence, and a paragraph; performing part-of-speech analysis on the sub-text to obtain some first words whose part of speech is noun in the sub-text.

[0040] In the above optional embodiment, the position of the first word in the original text is first located, that is, the fragment text (i.e., sub-text) where the first word is located is searched, and then the part of speech of the extracted first word is determined. When the first word is a noun, some of the first words in the sub-text whose part of speech is noun are extracted.

[0041] It should be noted that subtext includes but is not limited to: phrases, sentences and paragraphs.

[0042] As an optional embodiment, after obtaining some first words whose part of speech is noun in the sub-text, the post-processing method of speech recognition further includes: adding a tag to the sub-text.

[0043] In the above optional embodiment, after extracting the first words of the plurality of first words, some of which are nouns, they are marked to facilitate the next step of processing.

[0044] As an optional embodiment, the post-processing method of speech recognition further includes: when the analysis result indicates that the first words in the sub-texts are all the same, releasing the mark of the sub-text.

[0045] In the above optional embodiment, if the analysis result indicates that the first words in multiple sub-texts are the same, there is no ambiguity and the mark is released; if the analysis result indicates that the first words in multiple sub-texts are different, the mark is retained.

[0046] As an optional embodiment, determining a second word to replace a first word in a subtext includes: processing part of the first word based on a predetermined algorithm to obtain the second word, wherein the predetermined algorithm includes at least one of the following: a probability analysis algorithm and a voting selection algorithm.

[0047] In the above optional embodiment, when it is determined that the extracted first word may be wrong, a context-based probability analysis algorithm is used to analyze multiple candidate words, and then a similarity voting algorithm is used to vote for the second word with the highest accuracy, and the second word is used to replace the first word, thereby improving the accuracy of speech recognition.

[0048] As can be seen from the above, in an embodiment of the present invention, the target text is firstly subjected to phoneme analysis to obtain multiple first words with the same phonemes in the target text; then, some of the first words whose part of speech in the target text is noun can be determined; then, the sub-text where some of the first words are located can be analyzed to obtain an analysis result; then, when the analysis result indicates that the first words in the sub-text are different, a second word that replaces the first word in the sub-text can be determined; finally, the first word in the sub-text can be replaced by the second word, thereby achieving the purpose of extracting words with the same pronunciation but different characters by utilizing the characteristics of the same phonemes and correcting the text by using text analysis, thereby achieving the technical effect of improving the accuracy of speech recognition, and further solving the technical problem of insufficient recognition of words with the same phonemes but different contents in the speech recognition process in the related art.

[0049] In addition, it should be noted that the post-processing method of speech recognition provided by the embodiment of the present invention can be applied to non-real-time recording recognition scenarios, which is more tolerant than the application scenarios of real-time speech recognition, and allows full-text noun analysis optimization after all speech recognition results are obtained, so as to achieve a higher recognition rate and more accurate semantic expression. The processing process is to transcribe the audio into text through speech recognition technology, and perform contextual semantic association optimization processing on the transcribed results at the chapter level. The processing process includes: same phoneme recognition, part of speech analysis, hit word selection, and expression unification.

[0050] That is, the first step, speech recognition, uses an acoustic model to convert sound signals into phonemes, and then decodes the phonemes into text information through a language model.

[0051] The second step is to analyze the same phonemes, using the phoneme information generated in the first step to find the same phonemes, such as the example "l i3yi1yi1" (candidate words: Li Yiyi, Li Yiyi) in the above text.

[0052] The third step is part-of-speech analysis. Find the text corresponding to the same phoneme as "l i3yi1yi1" from all contexts. After obtaining the corresponding text, analyze its part of speech in the sentence. If they are all nouns, mark the text for the next step of processing.

[0053] Step 4: Hit word selection. From the text marked in the previous step, a group of data is selected. If the texts in the group are the same, the semantic expression of the group of data is unambiguous and the mark can be released directly. If the texts in the group are different, the group of data is taken out for probability analysis and voting. The selection of hit words can use the simplest word frequency as a voting indicator, and select the most frequently appearing words as hit words. You can also use other words in the entire text to calculate the semantic relevance to the hit word.

[0054] The fifth step is to unify the expression and replace the candidate words hit in the fourth step one by one to the position of the original candidate words. Through the above-mentioned post-processing method of speech recognition, after the speech recognition is completed, especially in the non-real-time speech recognition scenario, there is enough context information, and combined with the context information, the paragraph level optimization is performed. The invention can combine the context phonemes and word parts of speech for analysis, and uniformly process the nouns with the same phonemes to ensure the accuracy of the semantics of the noun in the context.

[0055] Example 2

[0056] According to another aspect of the embodiment of the present invention, a post-processing device for speech recognition is also provided. Figure 2 is a schematic diagram of a post-processing device for speech recognition according to an embodiment of the present invention, such as Figure 2 As shown, it includes: a first analyzing module 21, a first determining module 23, a second analyzing module 25, a second determining module 27 and a replacing module 29. The post-processing device of speech recognition is described below.

[0057] A first analysis module 21 is used to perform phoneme analysis on the target text to obtain a plurality of first words with the same phonemes in the target text;

[0058] A first determination module 23, configured to determine some of the first words whose part of speech in the target text is noun;

[0059] The second analysis module 25 is used to analyze the subtext where part of the first word is located to obtain an analysis result;

[0060] A second determination module 27, configured to determine a second word in the subtext to replace the first word when the analysis result indicates that the first word in the subtext is different;

[0061] The replacement module 29 is used to replace the first word in the subtext with the second word.

[0062] It should be noted that the first analysis module 21, the first determination module 23, the second analysis module 25, the second determination module 27 and the replacement module 29 correspond to steps S102 to S110 in Embodiment 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Embodiment 1. It should be noted that the above modules, as part of the device, can be executed in a computer system such as a set of computer executable instructions.

[0063] As can be seen from the above, in the embodiment of the present invention, the first analysis module 21 can be used to perform phoneme analysis on the target text to obtain multiple first words with the same phonemes in the target text; then the first determination module 23 is used to determine some of the first words whose part of speech in the target text is a noun; then the second analysis module 25 is used to analyze the subtext where some of the first words are located to obtain the analysis result; then the second determination module 27 is used to determine the second word in the subtext to replace the first word when the analysis result indicates that the first words in the subtext are different; finally, the replacement module 29 can be used to replace the first word in the subtext with the second word. The post-processing device for speech recognition provided by the embodiment of the present invention achieves the purpose of extracting words with the same pronunciation but different characters by using the characteristics of the same phonemes and correcting the text by using text analysis, thereby achieving the technical effect of improving the accuracy of speech recognition, and further solving the technical problem of insufficient recognition of words with the same phonemes but different contents in the speech recognition process in the related art.

[0064] Optionally, the post-processing device for speech recognition also includes: a conversion module, which is used to convert the original sound signal into phoneme information through an acoustic model before performing phoneme analysis on the target text; and a decoding module, which is used to decode the phoneme information using a language model to obtain the target text.

[0065] Optionally, the first determination module includes: a positioning unit, used to locate sub-texts where multiple first words are located in the target text, wherein the sub-text includes at least one of the following: a phrase, a sentence and a paragraph; an analysis unit, used to perform part-of-speech analysis on the sub-text to obtain some first words in the sub-text whose part of speech is noun.

[0066] Optionally, the speech recognition post-processing device further includes: a marking module, which is used to add a mark to the sub-text after obtaining the first part of the words whose part of speech is noun in the sub-text.

[0067] Optionally, the speech recognition post-processing device further comprises: a releasing module, configured to release the mark of the sub-text when the analysis result indicates that the first words in the sub-texts are all the same.

[0068] Optionally, the second determination module includes: a processing unit, configured to process part of the first words based on a predetermined algorithm to obtain the second words, wherein the predetermined algorithm includes at least one of the following: a probability analysis algorithm and a voting selection algorithm.

[0069] Example 3

[0070] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein when the computer program is executed by a processor, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned speech recognition post-processing methods.

[0071] Example 4

[0072] According to another aspect of an embodiment of the present invention, a processor is provided, and the processor is used to run a computer program, wherein the computer program executes any one of the above-mentioned post-processing methods for speech recognition when running.

[0073] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0074] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0076] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0077] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0078] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0079] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A post-processing method for speech recognition, characterized in that: include: Performing phoneme analysis on the target text to obtain a plurality of first words having the same phonemes in the target text; Determine that the part of speech of the plurality of first words in the target text is some of the first words of noun; Analyze the subtext where the first word of the part is located to obtain an analysis result; When the analysis result indicates that the first words in the subtext are different, determining a second word in the subtext to replace the first word; The first word in the subtext is replaced with the second word.

2. The method according to claim 1, characterized in that Before performing phoneme analysis on the target text, the method further includes: Convert the original sound signal into phoneme information through the acoustic model; The phoneme information is decoded using a language model to obtain the target text.

3. The method according to claim 1, characterized in that Determining some of the first words whose part of speech of the plurality of first words in the target text is a noun comprises: Locating the subtexts where the first words are located in the target text, wherein the subtexts include at least one of the following: phrases, sentences, and paragraphs; Part-of-speech analysis is performed on the sub-text to obtain the first words of the part of speech of the sub-text that are nouns.

4. The method according to claim 1, characterized in that: After obtaining the first part of the words whose part of speech is noun in the sub-text, the method further includes: adding a tag to the sub-text.

5. The method according to claim 4, characterized in that The method further comprises: When the analysis result indicates that the first words in the sub-texts are all the same, the mark of the sub-text is released.

6. The method according to claim 4, characterized in that Determining a second word in the subtext to replace the first word includes: The part of the first words is processed based on a predetermined algorithm to obtain the second words, wherein the predetermined algorithm includes at least one of the following: a probability analysis algorithm and a voting selection algorithm.

7. A post-processing device for speech recognition, characterized in that: include: A first analysis module is used to perform phoneme analysis on the target text to obtain a plurality of first words with the same phonemes in the target text; A first determination module is used to determine some of the first words of the plurality of first words whose part of speech in the target text is a noun; A second analysis module is used to analyze the subtext where the first word is located to obtain an analysis result; A second determination module, configured to determine a second word in the subtext to replace the first word when the analysis result indicates that the first word in the subtext is different; A replacement module is used to replace the first word in the subtext with the second word.

8. The device according to claim 7, characterized in that The device also includes: A conversion module, used for converting the original sound signal into phoneme information through an acoustic model before performing phoneme analysis on the target text; A decoding module is used to decode the phoneme information using a language model to obtain the target text.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed by a processor, the device where the computer-readable storage medium is located is controlled to execute the post-processing method of speech recognition described in any one of claims 1 to 6.

10. A processor, characterized in that: The processor is used to run a computer program, wherein the computer program executes the post-processing method of speech recognition according to any one of claims 1 to 6 when running.

Images