Method for reducing near-end speech misrecognition, speech recognition method and apparatus therefor

Abstract

The application provides a method for reducing similar speech misrecognition, a speech recognition method and a device thereof, wherein the method for reducing similar speech misrecognition comprises: obtaining similar speech misrecognized as keyword speech; determining a phoneme sequence of the similar speech and a phoneme sequence of the keyword speech according to the similar speech and the keyword speech; comparing the phoneme sequence of the similar speech and the phoneme sequence of the keyword speech to determine a misrecognized phoneme sequence in the similar speech; adding at least one new mapping group in a pronunciation dictionary according to a phoneme replacement scheme; characters in the new mapping group include misrecognized characters, and a phoneme sequence in the new mapping group includes a phoneme sequence after at least one error-prone phoneme in the misrecognized phoneme sequence is removed, and the error-prone phoneme is a phoneme previously specified in the misrecognized phoneme sequence. The application can conveniently and effectively reduce the occurrence probability of similar speech misrecognition in speech recognition.

Description

Technical Field

[0001] This invention relates to the field of speech recognition technology, and in particular to a method, speech recognition method and apparatus for reducing misidentification of similar speech. Background Technology

[0002] The misidentification of similar speech involved in this invention refers to the situation in keyword speech recognition where some speech sounds similar to the keyword speech are misidentified as the keyword speech. For example, the keyword speech is "ABD", while a similar speech sound, i.e., a near-similarity speech sound, is "ABCD", where A, B, C, and D can be any character, such as a Chinese character or a word; C is the misidentified character. During the speech recognition process, "ABCD" will be misidentified as "ABD". Such misidentification is a difficult situation to handle in keyword speech recognition.

[0003] The most direct and effective solution to this problem is usually to significantly increase the amount of training data, enabling the speech model to have sufficient ability to recognize similar speech that has been misidentified. However, this approach requires additional manpower and is inconvenient to process. Furthermore, if the quantity and diversity of the training data are insufficient, it may not achieve the desired result.

[0004] Another solution to this problem is to adjust the weights of keywords and non-keywords in the speech recognition system's output, increasing the weight of non-keyword recognition results. Obviously, this method can reduce the probability of non-keywords, including similar words, being misidentified as keywords, i.e., reduce the probability of similar speech being misidentified as keyword speech. However, this comes at the cost of reducing the probability of keywords being correctly recognized, which is detrimental to the speech recognition system's ability to recognize keyword speech.

[0005] Therefore, how to conveniently and effectively reduce the probability of misrecognition of similar speech in speech recognition has become an urgent problem to be solved. Summary of the Invention

[0006] To address the aforementioned problems, the present invention provides a method, a speech recognition method, and an apparatus for reducing misidentification of similar speech. By adding new mapping groups to the pronunciation dictionary, the probability of misidentified characters being accurately recognized in speech recognition is increased, thereby conveniently and effectively reducing the probability of misidentification of similar speech.

[0007] In a first aspect, the present invention provides a method for reducing misidentification of similar speech, comprising:

[0008] Obtain similar speech that has been misidentified as keyword speech;

[0009] Determine the phoneme sequence of similar speech based on similar speech sounds, and determine the phoneme sequence of keyword speech based on keyword speech sounds;

[0010] By comparing the phoneme sequences of similar speech with the phoneme sequences of keyword speech, misidentified phoneme sequences in similar speech are identified. The misidentified phoneme sequence is the phoneme sequence corresponding to the misidentified character, and the misidentified character is the character that was misidentified in the character sequence corresponding to the similar speech.

[0011] Based on the phoneme replacement scheme, at least one new mapping group is added to the pronunciation dictionary, so that the speech recognition system can recognize the speech to be recognized based on the modified pronunciation dictionary;

[0012] The mapping group includes a character and at least one phoneme sequence. The speech recognition system is used to determine the character corresponding to the phoneme sequence in the mapping group of the pronunciation dictionary based on the phoneme sequence to be recognized. The phoneme sequence to be recognized is the phoneme sequence recognized by the speech recognition system from the speech to be recognized. The speech to be recognized includes similar speech. The characters in the newly added mapping group include misrecognized characters. The phoneme sequence in the newly added mapping group includes the phoneme sequence after removing at least one error-prone phoneme from the error-recognized phoneme sequence. The error-prone phoneme is a phoneme pre-specified in the error-recognized phoneme sequence.

[0013] Optionally, the phoneme replacement scheme includes: an inward-facing phoneme replacement scheme and a combined phoneme replacement scheme;

[0014] The steps for adding at least one new mapping group to the pronunciation dictionary based on the phoneme substitution scheme include:

[0015] According to the inward-facing phoneme replacement scheme, at least one misidentified phoneme in the misidentified phoneme sequence is removed; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the misidentified character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0016] Or / and, according to the combined phoneme replacement scheme, remove at least one misidentified phoneme from the misidentified phoneme sequence, modify the misidentified character into a combined character, which is a new character composed of the misidentified character and the characters adjacent to the misidentified character; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the combined character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0017] Optionally, the step of adding at least one new mapping group to the pronunciation dictionary according to the phoneme substitution scheme includes:

[0018] The phoneme replacement scheme is selected based on the number of phonemes in the misidentified phoneme sequence and the size of a specified threshold. The phoneme replacement schemes include: inward-type phoneme replacement scheme and joint-type phoneme replacement scheme.

[0019] When the number of phonemes in the misidentified phoneme sequence is greater than or equal to a specified threshold, at least one misidentified phoneme in the misidentified phoneme sequence is removed according to the inward-type phoneme replacement scheme; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the misidentified character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0020] When the number of phonemes in the misidentified phoneme sequence is less than a specified threshold, according to the combined phoneme replacement scheme, at least one misidentified phoneme in the misidentified phoneme sequence is removed, and the misidentified character is modified into a combined character, which is a new character composed of the misidentified character and the character adjacent to the misidentified character; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the combined character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0021] Secondly, the present invention provides a speech recognition method, the method comprising:

[0022] Acquire the speech to be recognized;

[0023] The modified pronunciation dictionary from any of the methods described above is used to perform speech recognition on the speech to be recognized, and the recognition result is output.

[0024] Thirdly, the present invention provides an apparatus for reducing misidentification of similar speech, comprising:

[0025] The first acquisition module is used to acquire similar speech that has been misidentified as keyword speech;

[0026] The determination module is used to determine the phoneme sequence of similar speech based on similar speech, and to determine the phoneme sequence of keyword speech based on keyword speech;

[0027] The comparison module is used to compare the phoneme sequences of similar speech and the phoneme sequences of keyword speech to identify misidentified phoneme sequences in similar speech. The misidentified phoneme sequence is the phoneme sequence corresponding to the misidentified character, and the misidentified character is the character that was misidentified in the character sequence corresponding to the similar speech.

[0028] The modification module is used to add at least one new mapping group to the pronunciation dictionary according to the phoneme replacement scheme, so that the speech recognition system can recognize the speech to be recognized according to the modified pronunciation dictionary.

[0029] The mapping group includes a character and at least one phoneme sequence. The speech recognition system is used to determine the character corresponding to the phoneme sequence in the mapping group of the pronunciation dictionary based on the phoneme sequence to be recognized. The phoneme sequence to be recognized is the phoneme sequence recognized by the speech recognition system from the speech to be recognized. The speech to be recognized includes similar speech. The characters in the newly added mapping group include misrecognized characters. The phoneme sequence in the newly added mapping group includes the phoneme sequence after removing at least one error-prone phoneme from the error-recognized phoneme sequence. The error-prone phoneme is a phoneme pre-specified in the error-recognized phoneme sequence.

[0030] Optionally, the phoneme replacement scheme includes: an inward-facing phoneme replacement scheme and a combined phoneme replacement scheme;

[0031] The modification module is also used to: remove at least one misidentified phoneme from the misidentified phoneme sequence according to the inward-facing phoneme replacement scheme; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the misidentified character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary;

[0032] Or / and, according to the combined phoneme replacement scheme, remove at least one misidentified phoneme from the misidentified phoneme sequence, modify the misidentified character into a combined character, which is a new character composed of the misidentified character and the characters adjacent to the misidentified character; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the combined character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0033] Optionally, the modified module includes:

[0034] The selection submodule is used to select a phoneme replacement scheme based on the number of phonemes in the misidentified phoneme sequence and the size of a specified threshold. The phoneme replacement schemes include: inward-type phoneme replacement schemes and joint-type phoneme replacement schemes.

[0035] The first modification submodule is used to remove at least one misidentified phoneme from the misidentified phoneme sequence according to the inward-type phoneme replacement scheme when the selection module determines that the number of phonemes in the misidentified phoneme sequence is greater than or equal to a specified threshold; and to use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group and the misidentified character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0036] The second modification submodule is used to remove at least one error-prone phoneme from the misidentified phoneme sequence according to the joint phoneme replacement scheme when the selection module determines that the number of phonemes in the misidentified phoneme sequence is less than a specified threshold. The misidentified character is modified into a combined character, which is a new character composed of the misidentified character and the character adjacent to the misidentified character. The modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the combined character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0037] Fourthly, the present invention provides a voice recognition device, the device comprising:

[0038] The second acquisition module is used to acquire the speech to be recognized;

[0039] The recognition module is used to perform speech recognition on the speech to be recognized using the modified pronunciation dictionary in any of the above methods for reducing misrecognition of similar speech, and output the recognition result.

[0040] Fifthly, the present invention provides an electronic device, the electronic device comprising:

[0041] At least one processor; and

[0042] A memory that is communicatively connected to at least one processor; wherein,

[0043] The memory stores instructions that can be executed by at least one processor to enable the at least one processor to perform any of the above methods for reducing misrecognition of similar speech.

[0044] Sixthly, the present invention provides an electronic device, the electronic device comprising:

[0045] At least one processor; and

[0046] A memory that is communicatively connected to at least one processor; wherein,

[0047] The memory stores instructions that can be executed by at least one processor, which enables the at least one processor to perform the speech recognition method described above.

[0048] In a seventh aspect, the present invention provides a chip, the chip comprising:

[0049] At least one processor; and

[0050] A memory that is communicatively connected to at least one processor; wherein,

[0051] The memory stores instructions that can be executed by at least one processor to enable the at least one processor to perform any of the above methods for reducing misrecognition of similar speech.

[0052] Eighthly, the present invention provides a chip, the chip comprising:

[0053] At least one processor; and

[0054] A memory that is communicatively connected to at least one processor; wherein,

[0055] The memory stores instructions that can be executed by at least one processor, which enables the at least one processor to perform the speech recognition method described above.

[0056] In a ninth aspect, the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the method for reducing misrecognition of similar speech as described above.

[0057] In a tenth aspect, the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the speech recognition method described above.

[0058] The method, speech recognition method and apparatus for reducing misidentification of similar speech provided in the embodiments of the present invention increase the accuracy of most phonemes corresponding to misidentified characters by adding at least one new mapping group to the pronunciation dictionary, thereby enabling quick and accurate matching of misidentified characters and improving the probability of accurate identification of misidentified characters in speech recognition, thus conveniently and effectively reducing the probability of misidentification of similar speech. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a schematic flowchart illustrating a method for reducing misidentification of similar speech according to an embodiment of this application;

[0061] Figure 2 This is a schematic flowchart illustrating a method for reducing misidentification of similar speech according to an embodiment of this application;

[0062] Figure 3 This is a schematic flowchart illustrating a method for reducing misidentification of similar speech according to an embodiment of this application;

[0063] Figure 4This is a schematic flowchart illustrating a speech recognition method according to an embodiment of this application;

[0064] Figure 5 This is a schematic structural diagram of a device for reducing misrecognition of similar speech according to an embodiment of this application;

[0065] Figure 6 This is a schematic structural diagram of a voice recognition device according to an embodiment of this application. Detailed Implementation

[0066] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0067] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0068] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising,” “including,” or “having,” etc., specify the presence of the stated feature, whole, step, operation, component, part, or combination thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof.

[0069] It should be noted that, in the speech recognition system mentioned in this embodiment, the speech features are first extracted from the speech to be recognized. Then, the speech features are processed by an acoustic model, an HMM (Hidden Markov Model), and a context switcher to output a corresponding phoneme sequence. Next, based on the mapping relationship between characters and phoneme sequences recorded in the pronunciation dictionary, the phoneme sequence is processed by a pronunciation dictionary transformation graph to obtain the corresponding characters. Finally, the character sequence is normalized by a language model to obtain the final recognition result. This application performs phoneme replacement at the pronunciation dictionary transformation graph, thereby reducing the probability of misrecognition of similar sounds.

[0070] Furthermore, taking the case where "ABCD" is misidentified as "ABD" as an example, let's explain the specific situation of speech misidentification: First, during the training of the acoustic model, the speech features of the phrase "ABCD" are mapped by the acoustic model to the phoneme sequence "A1 A2 A3 B1B2 B3 C1 C2 C3 D1 D2 D3". However, in the application stage, the speech features of "ABCD" are mapped to the phoneme sequence "A1A2 A3 B1 B2 B3 D1 D2 D3". Analyzing the misidentification process from a theoretical perspective, we can see that the output of the acoustic model corresponds to the transition mode of the Hidden Memory Model (HMM). The hidden states of the HMM represent phonemes. Here, the speech frame originally belonging to "C1 C2 C3" is incorrectly mapped to the HMM state of the adjacent preceding phoneme "B3" or following phoneme "D1". Clearly, this is because the speech features originally belonging to "C1 C2 C3" do not have a strong correspondence with the HMM state corresponding to "C1 C2 C3". This could be due to insufficient modeling in the model itself, or it could be due to the inherent ambiguity of speech. For example, the misidentified character "C" is a transient sound, so the pronunciation of "C" may be difficult to distinguish from the pronunciation of the characters before and after it. The speech features of "C" are then mapped to the HMM transfer network of the characters before and after it.

[0071] Regarding the above situation, the phoneme substitution theory in this application makes the following assumptions: 1) In a sequence of misidentified phonemes corresponding to a misidentified character, it is not that the speech features corresponding to the misidentified character have no strong correspondence with each phoneme in the sequence. 2) The speech features extracted from a continuous speech segment corresponding to a misidentified character have a higher probability of corresponding to the phoneme of a character adjacent to the misidentified character than to an irrelevant phoneme. Irrelevant phonemes are those corresponding to characters not adjacent to the misidentified character, and are different from the phonemes corresponding to characters adjacent to the misidentified character.

[0072] Based on the above two assumptions, this application adds "C1 C2" and / or "C2 C3" to the original "C1 C2 C3" phoneme sequence corresponding to the character "C" in the pronunciation dictionary. This method of adding a shorter corresponding phoneme sequence to the character "C" based on its original phoneme sequence is defined as an inward-facing phoneme replacement scheme. Taking "C1 C2" as an example, it is assumed that of the three phonemes in the original phoneme sequence "C1 C2 C3", only "C3" has a weak correspondence with the audio features, while "C1" and "C2" have strong correspondences with their respective audio features. Replacing the original phoneme sequence "C1 C2 C3" of the character "C" in the pronunciation dictionary with "C1 C2" essentially maps the audio features of "C3" to the HMM network of "C2". Of course, it is possible that the audio features of "C3" are mapped to the HMM network of "D1". Compared to existing methods that incorrectly assign all three phonemes "C1 C2 C3" to the character "C", this inward-facing phoneme replacement scheme has a higher probability of making the audio features of misidentified characters in the speech to be recognized correspond more strongly to the new phoneme sequence "C1 C2", thus correctly decoding the character "C".

[0073] Similarly, assuming that the correspondence between "C1" and the audio features is weak, while the correspondences between "C2" and "C3" and their respective audio features are strong, and that the feature of "C1" can correspond to the HMM network of the adjacent "B3" on the left, this application can further add "B1 B2 B3 C2 C3" as the phoneme sequence corresponding to the new character "BC" based on the phoneme sequence corresponding to the character "C" in the original "C1 C2 C3" and the position of the misidentified phoneme sequence in the phoneme sequence corresponding to the entire speech to be recognized. This increases the probability of correctly decoding the two "BC" characters in the character sequence corresponding to the speech to be recognized. Here, this scheme of shortening the misidentified phoneme sequence of a character and then merging it with adjacent characters, so that multiple characters can be recognized at once from the merged phoneme sequence, is defined as a joint phoneme replacement scheme.

[0074] Thus, when recognizing similar speech for the first time using the scheme provided in this application, the phoneme sequence corresponding to the misidentified character can be added through the inward-facing phoneme replacement scheme and / or the combined phoneme replacement scheme. Subsequently, when recognizing the same similar speech, the corresponding character sequence can be accurately identified without modifying the pronunciation dictionary. Furthermore, considering the size of the pronunciation dictionary, one of the schemes—the inward-facing phoneme replacement scheme and the combined phoneme replacement scheme—can be selected to modify the pronunciation dictionary based on its effectiveness in successfully recognizing misidentified characters.

[0075] Firstly, this embodiment provides a method for reducing misidentification of similar speech, see [link to previous document]. Figure 1 The method includes steps S101 to S104:

[0076] Step S101: Obtain similar speech that was misidentified as keyword speech.

[0077] In one optional embodiment, the step of obtaining similar speech that has been misidentified as keyword speech includes: analyzing the recognition results of the speech recognition system and filtering similar speech from the speech recognized by the speech recognition system. Further, similar speech with more than a predetermined number of misidentifications can be processed as in this embodiment; however, in this optional embodiment, the predetermined value is not specifically limited.

[0078] Step S102: Determine the phoneme sequence of similar speech based on similar speech, and determine the phoneme sequence of keyword speech based on keyword speech.

[0079] Step S103: Compare the phoneme sequences of similar speech with the phoneme sequences of keyword speech to identify misidentified phoneme sequences in similar speech.

[0080] Among them, the misidentified phoneme sequence is the phoneme sequence corresponding to the misidentified character, and the misidentified character is the character that was misidentified in the character sequence corresponding to similar speech.

[0081] For example, comparing the phoneme sequence "A1 A2 A3 B1 B2 B3 C1 C2 C3 D1 D2 D3" of similar speech with the phoneme sequence "A1 A2 A3 B1 B2 B3 D1 D2 D3" of keyword speech, we can determine that "C1 C2 C3" is a misidentified phoneme sequence. Among them, the misidentified character corresponding to "C1 C2 C3" is "C".

[0082] Step S104: Add at least one new mapping group to the pronunciation dictionary according to the phoneme replacement scheme, so that the speech recognition system can recognize the speech to be recognized according to the modified pronunciation dictionary.

[0083] The mapping group includes: a character and at least a phoneme sequence; the speech recognition system is used to determine the character corresponding to the phoneme sequence to be recognized in the mapping group of the pronunciation dictionary; the phoneme sequence to be recognized is the phoneme sequence recognized by the speech recognition system from the speech to be recognized; the speech to be recognized includes similar speech; the characters in the newly added mapping group include misrecognized characters; the phoneme sequence in the newly added mapping group includes the phoneme sequence after removing at least one error-prone phoneme from the error-recognized phoneme sequence; the error-prone phoneme is a phoneme pre-specified in the error-recognized phoneme sequence.

[0084] It should be noted that the factors for identifying misidentified phonemes in misidentified characters include, but are not limited to, the pronunciation rules of the misidentified characters or the pronunciation differences between the misidentified phoneme sequence and the adjacent phoneme sequences in the corresponding whole phoneme sequence of the speech. For example, consonants in the misidentified phoneme sequence can be considered misidentified phonemes. Specifically, a trained neural network model can also be used for differentiation; this embodiment does not impose specific limitations. By adding new mapping groups without deleting the original mapping groups that map misidentified factor sequences to misidentified characters, it can be ensured that misidentified characters are efficiently and accurately identified when they do not appear as misidentified characters in other speech.

[0085] In one optional embodiment, the phoneme replacement scheme includes: an inward-facing phoneme replacement scheme and a combined phoneme replacement scheme;

[0086] The steps for adding at least one new mapping group to the pronunciation dictionary based on the phoneme substitution scheme include:

[0087] According to the inward-facing phoneme replacement scheme, at least one misidentified phoneme in the misidentified phoneme sequence is removed; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the misidentified character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0088] Or / and, according to the combined phoneme replacement scheme, remove at least one misidentified phoneme from the misidentified phoneme sequence, modify the misidentified character into a combined character, which is a new character composed of the misidentified character and the characters adjacent to the misidentified character; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the combined character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0089] In this case, the misidentified phoneme sequence may contain multiple error-prone phonemes. When adding a new mapping group, one of these error-prone phonemes is selected and removed, resulting in two mapping groups that can be added to the pronunciation dictionary. For example, in the misidentified phoneme sequence "C1 C2 C3", C1 and C3 are both error-prone phonemes. According to the inward-facing phoneme replacement scheme, a mapping group can be added to the pronunciation dictionary that maps the misidentified character "C" to the new phoneme sequence "C2C3" and another that maps the misidentified character "C" to the new phoneme sequence "C1 C2". Which new mapping group to add to the pronunciation dictionary, or whether to add both, can be chosen based on the actual situation; this embodiment does not impose a specific limitation.

[0090] Similarly, in the misidentified phoneme sequence "C1 C2 C3", C1 and C3 are both error-prone phonemes. In this case, according to the combined phoneme replacement scheme, mapping groups can be added to the pronunciation dictionary to map the misidentified character "BC" to the new phoneme sequence "B1 B2 B3C2 C3" and to map the misidentified character "BC" to the new phoneme sequence "C1 C2D1 D2 D3". Which new mapping group to add to the pronunciation dictionary, or whether to add both new mapping groups, can be chosen based on the actual situation; this embodiment does not impose specific limitations.

[0091] In one alternative embodiment, the step of adding at least one new mapping group to the pronunciation dictionary according to the phoneme substitution scheme includes:

[0092] The phoneme replacement scheme is selected based on the number of phonemes in the misidentified phoneme sequence and the value of a specified threshold. The phoneme replacement schemes include: inward-facing phoneme replacement schemes and combined phoneme replacement schemes.

[0093] When the number of phonemes in the misidentified phoneme sequence is greater than or equal to a specified threshold, at least one misidentified phoneme in the misidentified phoneme sequence is removed according to the inward-type phoneme replacement scheme; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the misidentified character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0094] When the number of phonemes in the misidentified phoneme sequence is less than a specified threshold, according to the combined phoneme replacement scheme, at least one misidentified phoneme in the misidentified phoneme sequence is removed, and the misidentified character is modified into a combined character, which is a new character composed of the misidentified character and the character adjacent to the misidentified character; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the combined character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0095] The specified threshold includes, but is not limited to, integers of three or greater than three. By modifying the pronunciation dictionary through a phoneme replacement scheme, the probability of misidentification of similar speech can be reduced, and the excessive complexity of processing phoneme sequences through the pronunciation dictionary transformation graph due to the large size of the pronunciation dictionary can also be avoided.

[0096] Secondly, this embodiment provides a method for reducing misidentification of similar speech, see [link to relevant documentation]. Figure 2 The method includes steps S201 to S204:

[0097] Step S201: Obtain similar speech that was misidentified as keyword speech.

[0098] Step S202: Determine the phoneme sequence of similar speech based on similar speech, and determine the phoneme sequence of keyword speech based on keyword speech.

[0099] Step S203: Compare the phoneme sequences of similar speech with the phoneme sequences of keyword speech to identify misidentified phoneme sequences in similar speech.

[0100] Step S204: According to the inward-type phoneme replacement scheme, remove at least one misidentified phoneme from the misidentified phoneme sequence; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the misidentified character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary; according to the combined-type phoneme replacement scheme, remove at least one misidentified phoneme from the misidentified phoneme sequence, modify the misidentified character into a combined character, the combined character being a new character composed of the misidentified character and the character adjacent to the misidentified character; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the combined character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0101] Thirdly, this embodiment provides a method for reducing misidentification of similar speech, see [link to relevant documentation]. Figure 3 The method includes steps S301 to S306:

[0102] Step S301: Obtain similar speech that was misidentified as keyword speech.

[0103] Step S302: Determine the phoneme sequence of similar speech based on similar speech, and determine the phoneme sequence of keyword speech based on keyword speech.

[0104] Step S303: Compare the phoneme sequences of similar speech with the phoneme sequences of keyword speech to identify misidentified phoneme sequences in similar speech.

[0105] Step S304: Determine the number of phonemes in the misidentified phoneme sequence and the size of the specified threshold.

[0106] Step S305: When the number of phonemes in the misidentified phoneme sequence is greater than or equal to a specified threshold, remove at least one misidentified phoneme from the misidentified phoneme sequence according to the inward-facing phoneme replacement scheme; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the misidentified character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0107] Step S306: When the number of phonemes in the misidentified phoneme sequence is less than a specified threshold, according to the combined phoneme replacement scheme, remove at least one misidentified phoneme from the misidentified phoneme sequence, modify the misidentified character into a combined character, which is a new character composed of the misidentified character and the character adjacent to the misidentified character; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the combined character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0108] Fourthly, this embodiment provides a speech recognition method, see [link to relevant documentation]. Figure 4 The method includes steps S401 to S402:

[0109] Step S401: Obtain the speech to be recognized.

[0110] Step S402: Perform speech recognition on the speech to be recognized using the modified pronunciation dictionary and speech recognition system in any of the above methods for reducing misrecognition of similar speech, and output the recognition result.

[0111] This speech recognition method uses a modified pronunciation dictionary to perform speech recognition, which can improve the probability of accurately recognizing misidentified characters in speech recognition, thereby conveniently and effectively reducing the probability of misrecognition of similar speech.

[0112] Fifthly, this embodiment provides a device 500 for reducing misrecognition of similar speech, see [link to previous document]. Figure 5 The device 500 for reducing misrecognition of similar speech includes:

[0113] The first acquisition module 501 is used to acquire similar speech that has been misidentified as keyword speech;

[0114] The determining module 502 is used to determine the phoneme sequence of similar speech based on similar speech, and to determine the phoneme sequence of keyword speech based on keyword speech;

[0115] The comparison module 503 is used to compare the phoneme sequence of similar speech and the phoneme sequence of keyword speech to determine the misidentified phoneme sequence in the similar speech. The misidentified phoneme sequence is the phoneme sequence corresponding to the misidentified character, and the misidentified character is the character that is misidentified in the character sequence corresponding to the similar speech.

[0116] Modify module 504 to add at least one new mapping group to the pronunciation dictionary according to the phoneme replacement scheme, so that the speech recognition system can recognize the speech to be recognized according to the modified pronunciation dictionary;

[0117] The mapping group includes a character and at least one phoneme sequence. The speech recognition system is used to determine the character corresponding to the phoneme sequence in the mapping group of the pronunciation dictionary based on the phoneme sequence to be recognized. The phoneme sequence to be recognized is the phoneme sequence recognized by the speech recognition system from the speech to be recognized. The speech to be recognized includes similar speech. The characters in the newly added mapping group include misrecognized characters. The phoneme sequence in the newly added mapping group includes the phoneme sequence after removing at least one error-prone phoneme from the error-recognized phoneme sequence. The error-prone phoneme is a phoneme pre-specified in the error-recognized phoneme sequence.

[0118] In one optional embodiment, the phoneme replacement scheme includes: an inward-facing phoneme replacement scheme and a combined phoneme replacement scheme;

[0119] Modification module 504 is also used to: remove at least one misidentified phoneme from the misidentified phoneme sequence according to the inward-type phoneme replacement scheme; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the misidentified character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary;

[0120] Or / and, according to the combined phoneme replacement scheme, remove at least one misidentified phoneme from the misidentified phoneme sequence, modify the misidentified character into a combined character, which is a new character composed of the misidentified character and the characters adjacent to the misidentified character; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the combined character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0121] In an optional embodiment, the modification module 504 includes:

[0122] The selection submodule is used to select a phoneme replacement scheme based on the number of phonemes in the misidentified phoneme sequence and the size of a specified threshold. The phoneme replacement schemes include: inward-type phoneme replacement schemes and joint-type phoneme replacement schemes.

[0123] The first modification submodule is used to remove at least one misidentified phoneme from the misidentified phoneme sequence according to the inward-type phoneme replacement scheme when the selection module determines that the number of phonemes in the misidentified phoneme sequence is greater than or equal to a specified threshold; and to use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group and the misidentified character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0124] The second modification submodule is used to remove at least one error-prone phoneme from the misidentified phoneme sequence according to the joint phoneme replacement scheme when the selection module determines that the number of phonemes in the misidentified phoneme sequence is less than a specified threshold. The misidentified character is modified into a combined character, which is a new character composed of the misidentified character and the character adjacent to the misidentified character. The modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the combined character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

[0125] Sixthly, this embodiment provides a voice recognition device 600, see [link to previous document]. Figure 6 The voice recognition device 600 includes:

[0126] The second acquisition module 601 is used to acquire the speech to be recognized;

[0127] The recognition module 602 is used to perform speech recognition on the speech to be recognized using the modified pronunciation dictionary and speech recognition system in any of the above methods for reducing misrecognition of similar speech, and to output the recognition result.

[0128] Seventhly, this embodiment provides an electronic device, which includes:

[0129] At least one processor; and

[0130] A memory that is communicatively connected to at least one processor; wherein,

[0131] The memory stores instructions that can be executed by at least one processor to enable the at least one processor to perform the method for reducing misrecognition of similar speech as described in any of the above methods.

[0132] Eighthly, this embodiment provides an electronic device, which includes:

[0133] At least one processor; and

[0134] A memory that is communicatively connected to at least one processor; wherein,

[0135] The memory stores instructions that can be executed by at least one processor, which enables the at least one processor to perform the speech recognition method described above.

[0136] Ninthly, this embodiment provides a chip, the chip comprising:

[0137] At least one processor; and

[0138] A memory that is communicatively connected to at least one processor; wherein,

[0139] The memory stores instructions that can be executed by at least one processor to enable the at least one processor to perform the method for reducing misrecognition of similar speech as described in any of the preceding aspects.

[0140] Tenthly, this embodiment provides a chip, the chip comprising:

[0141] At least one processor; and

[0142] A memory that is communicatively connected to at least one processor; wherein,

[0143] The memory stores instructions that can be executed by at least one processor, which enables the at least one processor to perform the speech recognition method described above.

[0144] Eleventhly, this embodiment provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the method for reducing misrecognition of similar speech as described in any of the preceding aspects.

[0145] In a twelfth aspect, this embodiment provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the speech recognition method described above.

[0146] In the description of this specification, the references to terms such as "some embodiments," "other embodiments," "ideal embodiments," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example that are included in at least one embodiment or example of this application. In this specification, the illustrative descriptions of the above terms do not necessarily refer to the same embodiments or examples.

[0147] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0148] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for reducing misidentification of similar speech, characterized in that, include: Obtain similar speech that has been misidentified as keyword speech; The phoneme sequence of the similar speech is determined based on the similar speech, and the phoneme sequence of the keyword speech is determined based on the keyword speech; The phoneme sequences of the similar speech and the phoneme sequences of the keyword speech are compared to determine the misidentified phoneme sequences in the similar speech. The misidentified phoneme sequences are the phoneme sequences corresponding to the misidentified characters, and the misidentified characters are the characters that are misidentified in the character sequences corresponding to the similar speech. According to the phoneme replacement scheme, at least one new mapping group is added to the pronunciation dictionary, so that the speech recognition system can recognize the speech to be recognized based on the modified pronunciation dictionary; The mapping group includes a character and at least one phoneme sequence. The speech recognition system determines the character corresponding to the phoneme sequence in the mapping group of the pronunciation dictionary based on the phoneme sequence to be recognized. The phoneme sequence to be recognized is the phoneme sequence recognized by the speech recognition system from the speech to be recognized. The speech to be recognized includes the similar speech. The character in the newly added mapping group includes the misrecognized character. The phoneme sequence in the newly added mapping group includes the phoneme sequence after removing at least one error-prone phoneme from the error-prone phoneme sequence. The error-prone phoneme is a phoneme pre-specified in the error-prone phoneme sequence.

2. The method according to claim 1, characterized in that, The phoneme replacement schemes include: inward-type phoneme replacement schemes and combined phoneme replacement schemes; The step of adding at least one new mapping group to the pronunciation dictionary according to the phoneme replacement scheme includes: According to the inward-facing phoneme replacement scheme, at least one misidentified phoneme in the misidentified phoneme sequence is removed; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the misidentified character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary; Alternatively, according to the combined phoneme replacement scheme, at least one misidentified phoneme in the misidentified phoneme sequence is removed, and the misidentified character is modified into a combined character, wherein the combined character is a new character composed of the misidentified character and the character adjacent to the misidentified character; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the combined character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

3. The method according to claim 1, characterized in that, The step of adding at least one new mapping group to the pronunciation dictionary according to the phoneme replacement scheme includes: The phoneme replacement scheme is selected based on the number of phonemes in the misidentified phoneme sequence and the size of a specified threshold. The phoneme replacement scheme includes: an inward-type phoneme replacement scheme and a combined phoneme replacement scheme. When the number of phonemes in the misidentified phoneme sequence is greater than or equal to a specified threshold, at least one misidentified phoneme in the misidentified phoneme sequence is removed according to the inward-type phoneme replacement scheme; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the misidentified character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary. When the number of phonemes in the misidentified phoneme sequence is less than a specified threshold, at least one misidentified phoneme in the misidentified phoneme sequence is removed according to the combined phoneme replacement scheme, and the misidentified character is modified into a combined character, which is a new character composed of the misidentified character and the character adjacent to the misidentified character; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the combined character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

4. A speech recognition method, characterized in that, The method includes: Acquire the speech to be recognized; The modified pronunciation dictionary as described in any one of claims 1 to 3 is used to perform speech recognition on the speech to be recognized, and the recognition result is output.

5. A device for reducing misrecognition of similar speech, characterized in that, include: The first acquisition module is used to acquire similar speech that has been misidentified as keyword speech; The determining module is used to determine the phoneme sequence of the similar speech based on the similar speech, and to determine the phoneme sequence of the keyword speech based on the keyword speech; The comparison module is used to compare the phoneme sequence of the similar speech and the phoneme sequence of the keyword speech to determine the misidentified phoneme sequence in the similar speech. The misidentified phoneme sequence is the phoneme sequence corresponding to the misidentified character, and the misidentified character is the character that was misidentified in the character sequence corresponding to the similar speech. The modification module is used to add at least one new mapping group to the pronunciation dictionary according to the phoneme replacement scheme, so that the speech recognition system can recognize the speech to be recognized according to the modified pronunciation dictionary. The mapping group includes a character and at least one phoneme sequence. The speech recognition system determines the character corresponding to the phoneme sequence in the mapping group of the pronunciation dictionary based on the phoneme sequence to be recognized. The phoneme sequence to be recognized is the phoneme sequence recognized by the speech recognition system from the speech to be recognized. The speech to be recognized includes the similar speech. The character in the newly added mapping group includes the misrecognized character. The phoneme sequence in the newly added mapping group includes the phoneme sequence after removing at least one error-prone phoneme from the error-prone phoneme sequence. The error-prone phoneme is a phoneme pre-specified in the error-prone phoneme sequence.

6. The apparatus according to claim 5, characterized in that, The phoneme replacement schemes include: inward-type phoneme replacement schemes and combined phoneme replacement schemes; The modification module is further configured to: remove at least one misidentified phoneme from the misidentified phoneme sequence according to the inward-facing phoneme replacement scheme; use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group, and use the misidentified character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary; Alternatively, according to the combined phoneme replacement scheme, at least one misidentified phoneme in the misidentified phoneme sequence is removed, and the misidentified character is modified into a combined character, wherein the combined character is a new character composed of the misidentified character and the character adjacent to the misidentified character; the modified misidentified phoneme sequence is used as the phoneme sequence in the new mapping group, and the combined character is used as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

7. The apparatus according to claim 5, characterized in that, The modification module includes: The selection submodule is used to select a phoneme replacement scheme based on the number of phonemes in the misidentified phoneme sequence and the size of a specified threshold. The phoneme replacement scheme includes: an inward-type phoneme replacement scheme and a combined phoneme replacement scheme. The first modification submodule is used to remove at least one misidentified phoneme from the misidentified phoneme sequence according to the inward-facing phoneme replacement scheme when the selection submodule determines that the number of phonemes in the misidentified phoneme sequence is greater than or equal to a specified threshold; and to use the modified misidentified phoneme sequence as the phoneme sequence in the new mapping group and the misidentified character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary. The second modification submodule is used to, when the selection submodule determines that the number of phonemes in the misidentified phoneme sequence is less than a specified threshold, remove at least one misidentified phoneme from the misidentified phoneme sequence according to a combined phoneme replacement scheme, modify the misidentified character into a combined character, wherein the combined character is a new character composed of the misidentified character and a character adjacent to the misidentified character; and use the modified misidentified phoneme sequence as the phoneme sequence in a new mapping group, and use the combined character as the character in the new mapping group, so as to add at least one new mapping group to the pronunciation dictionary.

8. A voice recognition device, characterized in that, The device includes: The second acquisition module is used to acquire the speech to be recognized; The recognition module is used to perform speech recognition on the speech to be recognized using the modified pronunciation dictionary in any one of claims 1 to 3, and output the recognition result.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1 to 3.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of claim 4.

11. A chip, characterized in that, The chip includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1 to 3.

12. A chip, characterized in that, The chip includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of claim 4.

13. A computer-readable storage medium, wherein, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the method as described in any one of claims 1 to 3.

14. A computer-readable storage medium, wherein, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the method as described in claim 4.

Images