A Chinese homophone recommendation method and device, a storage medium and an apparatus
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ANHUI IFLYTEK INTELLIGENT SYST
- Filing Date
- 2023-01-17
- Publication Date
- 2026-06-02
Smart Images
Figure CN116072126B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of speech processing technology, and in particular to a method, apparatus, storage medium and device for recommending identical Chinese pronunciations. Background Technology
[0002] With the continuous breakthroughs in artificial intelligence technology and the increasing popularity of various smart devices, voiceprint comparison and identification technology is now widely used in fields such as smart homes, finance, and public security. Voiceprint comparison and identification requires first identifying multiple sets of identical syllables or phonemes, then performing deviation analysis on the syllables or phonemes to calculate the deviation rate, and combining this with stability analysis and prosodic feature analysis of the syllables or phonemes to comprehensively determine whether the person being compared is the same person.
[0003] Among the methods and tools for voiceprint comparison and identification, finding identical sounds is the most crucial step. However, existing methods for finding identical sounds in Chinese mainly involve manually identifying identical syllables or phonemes, or searching for identical syllables or phonemes using a 1:1 or 1:N method. This results in current voiceprint identification methods relying heavily on the personal experience of the evaluators, requiring repeated listening and searching, which leads to lengthy voiceprint identification processes and poor operational convenience. Summary of the Invention
[0004] The main objective of this application is to provide a method, apparatus, storage medium, and device for recommending Chinese homophones, which can effectively improve the efficiency and convenience of voiceprint identification by quickly and accurately recommending word information corresponding to homophones during voiceprint comparison and identification.
[0005] This application provides a method for recommending homophones in Chinese, including:
[0006] Acquire sample speech and target speech to be identified;
[0007] Using a pre-built homonym database, syllable recognition is performed on the sample speech and the target speech to obtain the words contained in the sample speech and the target speech and the word frequency corresponding to each word;
[0008] Using the terms contained in the sample speech and the target speech, and the word frequencies corresponding to each term, a term matrix and triples are constructed;
[0009] The triples are sorted, and based on the processing results, the word information corresponding to the same syllable to be recommended is determined for voiceprint identification.
[0010] In one possible implementation, the step of using a pre-built homonym database to perform syllable recognition on the sample speech and the target speech to obtain the words contained in the sample speech and the target speech, and the word frequency corresponding to each word, includes:
[0011] Using a pre-built stop word lexicon, neutral tone word lexicon, and empirical homophone word lexicon, syllable recognition is performed on the sample speech and the target speech to obtain the words contained in the sample speech and the target speech and the word frequency corresponding to each word.
[0012] In one possible implementation, the step of using a pre-built stop word lexicon, neutral tone word lexicon, and empirical homophone lexicon to perform syllable recognition on the sample speech and the target speech to obtain the words contained in the sample speech and the target speech and the word frequency corresponding to each word includes:
[0013] Speech recognition is performed on the sample speech and the target speech respectively to obtain the sample syllable sequence and sample Chinese character document contained in the sample speech, and the target syllable sequence and target Chinese character document contained in the target speech;
[0014] Using the sample syllable sequence, the sample Chinese character document is segmented to obtain preliminary sample terms corresponding to the sample Chinese character document; and using the target syllable sequence, the target Chinese character document is segmented to obtain preliminary target terms corresponding to the target Chinese character document.
[0015] The sample preliminary terms are merged to obtain merged sample preliminary terms; and the target preliminary terms are merged to obtain merged target preliminary terms.
[0016] Using a pre-built stop word lexicon, a neutral tone word lexicon, and an empirical homonym lexicon, the word frequency corresponding to the preliminary sample words after merging is calculated, as well as the word frequency corresponding to the preliminary target words after merging is calculated.
[0017] In one possible implementation, the step of merging identical terms in the preliminary sample terms to obtain merged preliminary sample terms; and merging identical terms in the preliminary target terms to obtain merged preliminary target terms, includes:
[0018] The sample preliminary terms are merged with identical terms, and the processed sample preliminary terms are sorted in ascending alphabetical order to obtain merged sample preliminary terms; and the target preliminary terms are merged with identical terms, and the processed target preliminary terms are sorted in ascending alphabetical order to obtain merged target preliminary terms.
[0019] In one possible implementation, the step of calculating the word frequency corresponding to the preliminary entries of the merged sample using a pre-built stop word lexicon, a neutral tone word lexicon, and an empirical homonym lexicon includes:
[0020] Using a pre-built stop word lexicon, a neutral tone word lexicon, and an empirical homonym lexicon, determine whether the merged sample preliminary entries belong to the stop word lexicon, the neutral tone word lexicon, or the empirical homonym lexicon;
[0021] If so, the word frequency corresponding to the preliminary word entries of the merged sample is calculated by using the preset recommendation weights corresponding to stop words, unstressed words, or empirical homophones.
[0022] In one possible implementation, the step of constructing a word matrix and triples using the words contained in the sample speech and the target speech respectively includes:
[0023] Construct a word matrix by using the words contained in the sample speech as rows and the words contained in the target speech as columns;
[0024] Iterate through the word matrix sequentially and compare whether the words contained in the sample speech in the rows and the words contained in the target speech in the columns are the same;
[0025] If so, multiply the frequency of the same word in the sample speech with the frequency of the same word in the target speech, and use the product to construct a triplet.
[0026] In one possible implementation, the process of sorting the triples and determining the word information corresponding to the same syllable to be recommended based on the processing result for voiceprint identification includes:
[0027] The triples are sorted in descending order according to the product of word frequencies to obtain the sorting result;
[0028] The products of the first N word frequencies in the sorting results correspond to the same words, which are then used to determine the words corresponding to the same syllables to be recommended; where N is a positive integer greater than 0.
[0029] The start time, end time, and word frequency of the word corresponding to the same syllable to be recommended are determined in the sample speech and the target speech. The start time, end time, and word frequency are used to construct the word information corresponding to the same syllable to be recommended, which is then used for voiceprint identification.
[0030] This application also provides a Chinese homophone recommendation device, including:
[0031] The acquisition unit is used to acquire sample speech and target speech to be identified;
[0032] The recognition unit is used to perform syllable recognition on the sample speech and the target speech using a pre-built homophone database, so as to obtain the words contained in the sample speech and the target speech and the word frequency corresponding to each word;
[0033] The construction unit is used to construct a word matrix and triples by utilizing the words contained in the sample speech and the target speech and the word frequency corresponding to each word.
[0034] The determining unit is used to sort the triples and, based on the processing results, determine the word information corresponding to the same syllable to be recommended, for voiceprint identification.
[0035] In one possible implementation, the identification unit is specifically used for:
[0036] Using a pre-built stop word lexicon, neutral tone word lexicon, and empirical homophone word lexicon, syllable recognition is performed on the sample speech and the target speech to obtain the words contained in the sample speech and the target speech and the word frequency corresponding to each word.
[0037] In one possible implementation, the identification unit includes:
[0038] The recognition subunit is used to perform speech recognition on the sample speech and the target speech respectively, to obtain the sample syllable sequence and sample Chinese character document contained in the sample speech, and the target syllable sequence and target Chinese character document contained in the target speech;
[0039] The word segmentation subunit is used to segment the sample Chinese character document using the sample syllable sequence to obtain preliminary sample terms corresponding to the sample Chinese character document; and to segment the target Chinese character document using the target syllable sequence to obtain preliminary target terms corresponding to the target Chinese character document.
[0040] The merging subunit is used to perform the same term merging process on the sample preliminary terms to obtain the merged sample preliminary terms; and to perform the same term merging process on the target preliminary terms to obtain the merged target preliminary terms;
[0041] The first calculation subunit is used to calculate the word frequency corresponding to the preliminary sample word entries after merging, and to calculate the word frequency corresponding to the preliminary target word entries after merging, using a pre-built stop word lexicon, a neutral tone word lexicon, and an empirical homonym lexicon.
[0042] In one possible implementation, the merging subunit is specifically used for:
[0043] The sample preliminary terms are merged with identical terms, and the processed sample preliminary terms are sorted in ascending alphabetical order to obtain merged sample preliminary terms; and the target preliminary terms are merged with identical terms, and the processed target preliminary terms are sorted in ascending alphabetical order to obtain merged target preliminary terms.
[0044] In one possible implementation, the first computing subunit includes:
[0045] The judgment subunit is used to determine whether the merged sample preliminary entries belong to the stop word lexicon, the neutral tone word lexicon, or the empirical homonym lexicon by using a pre-built stop word lexicon, neutral tone word lexicon, and empirical homonym lexicon.
[0046] The second calculation subunit is used to calculate the word frequency corresponding to the preliminary term of the merged sample if it is determined that the preliminary term of the merged sample belongs to the stop word, neutral tone word or empirical homonym word library.
[0047] In one possible implementation, the building unit includes:
[0048] Construct sub-units to use the words contained in the sample speech as rows and the words contained in the target speech as columns to construct a word matrix;
[0049] The comparison subunit is used to sequentially traverse and compare whether the words contained in the sample speech in the rows of the word matrix and the words contained in the target speech in the columns are the same;
[0050] The multiplication subunit is used to multiply the frequency of the same word in the sample speech with the frequency of the same word in the target speech if the condition is met, and to construct a triple using the product result.
[0051] In one possible implementation, the determining unit includes:
[0052] The arrangement subunit is used to arrange the triples in descending order according to the product of word frequencies to obtain the arrangement result;
[0053] The subunit is used to determine the word corresponding to the same syllable to be recommended by multiplying the product of the N word frequencies in the sorting results; where N is a positive integer greater than 0.
[0054] The subunit is used to determine the start time, end time, and word frequency of the word corresponding to the same syllable to be recommended in the sample speech and the target speech, and to use the start time, end time, and word frequency to construct the word information corresponding to the same syllable to be recommended for voiceprint identification.
[0055] This application embodiment also provides a Chinese homophone recommendation device, including: a processor, a memory, and a system bus;
[0056] The processor and the memory are connected via the system bus;
[0057] The memory is used to store one or more programs, the one or more programs including instructions, which, when executed by the processor, cause the processor to perform any of the above-described methods for recommending Chinese homophones.
[0058] This application also provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform any of the above-described methods for recommending Chinese homophones.
[0059] This application also provides a computer program product, which, when run on a terminal device, causes the terminal device to execute any of the above-described methods for recommending Chinese homophones.
[0060] This application provides a method, apparatus, storage medium, and device for recommending homophones in Chinese. First, it acquires sample speech and target speech to be identified. Then, using a pre-constructed homophone dictionary, it performs syllable recognition on the sample and target speech to obtain the words contained in each speech and their corresponding word frequencies. Next, it constructs a word matrix and triples using the words and their corresponding word frequencies. The triples are then sorted, and based on the processing results, the word information corresponding to the homophones to be recommended is determined for subsequent voiceprint identification. Therefore, this application automatically determines the word information corresponding to the homophones to be recommended after performing syllable recognition on the sample and target speech using a pre-constructed homophone dictionary. This allows for faster and more accurate voiceprint comparison and identification of the user to whom the target speech belongs, effectively improving the efficiency and convenience of voiceprint identification. Attached Figure Description
[0061] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0062] Figure 1 A flowchart illustrating a method for recommending homophones in Chinese provided in this application embodiment;
[0063] Figure 2 A flowchart illustrating the overall process of recommending identical Chinese pronunciations provided in this application embodiment;
[0064] Figure 3 Example diagrams illustrating the specific structural composition of sample preliminary terms and target preliminary terms provided in the embodiments of this application;
[0065] Figure 4 An example diagram illustrating the specific compositional structure of the preliminary term "financial center" provided for embodiments of this application;
[0066] Figure 5 Another example diagram illustrating the specific compositional structure of the preliminary term "financial center" provided for embodiments of this application;
[0067] Figure 6 An example diagram showing word information corresponding to the same syllable to be recommended in the sample speech provided in the embodiments of this application;
[0068] Figure 7 An example diagram illustrating word information corresponding to the same syllable to be recommended in the target speech provided in this application embodiment;
[0069] Figure 8 This is a schematic diagram illustrating the composition of a Chinese homophone recommendation device provided in an embodiment of this application. Detailed Implementation
[0070] Currently, in voiceprint comparison and identification, a crucial step in determining whether someone is from the same person is to first identify identical sounds, and then use these identical sounds for subsequent voiceprint comparison and identification. Existing methods for finding identical sounds typically include three types: one is manually comparing and searching for identical syllables or phonemes, followed by deviation analysis, stability analysis, and prosodic feature analysis; another is searching for identical syllables or phonemes, followed by deviation analysis, stability analysis, and prosodic feature analysis; and the third is finding identical syllables or phonemes based on personal experience, followed by deviation analysis, stability analysis, and prosodic feature analysis.
[0071] It is evident that the existing methods of searching for identical sounds lead to the current voiceprint comparison and identification methods relying mainly on the personal experience of the identification personnel, requiring repeated listening and searching, which in turn makes voiceprint identification time-consuming and less convenient to operate.
[0072] To address the aforementioned shortcomings, this application provides a method for recommending homophones in Chinese. First, sample speech and target speech to be identified are acquired. Then, using a pre-constructed homophone lexicon, syllable recognition is performed on both the sample and target speech to obtain the words contained in each, along with their corresponding word frequencies. Next, a word matrix and triples are constructed using these words and their frequencies. The triples are then sorted, and based on the processing results, the word information corresponding to the homophones to be recommended is determined for subsequent voiceprint identification. Therefore, this application automatically determines the word information corresponding to the homophones to be recommended after syllable recognition of the sample and target speech using a pre-constructed homophone lexicon. This allows for faster and more accurate voiceprint comparison and identification of the user to whom the target speech belongs, effectively improving the efficiency and convenience of voiceprint identification.
[0073] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0074] First Embodiment
[0075] See Figure 1 This is a flowchart illustrating a method for recommending homophones in Chinese provided in this embodiment. The method includes the following steps:
[0076] S101: Obtain sample speech and target speech to be identified.
[0077] In this embodiment, any user requiring voiceprint comparison and identification is defined as the target user, the speech to be identified is defined as the target speech, and the speech already emitted by the target user as the identification benchmark is defined as the sample speech. It should be noted that this embodiment does not limit the length of the sample speech and the target speech; for example, both the sample speech and the target speech can be a sentence or a paragraph. Furthermore, this embodiment does not limit the acquisition scenario of the sample speech and the target speech; for example, both the sample speech and the target speech can be obtained in different scenarios such as a recording studio or a conference room.
[0078] For example, other voices spoken by the suspect at the request or guidance of the investigators can be used as sample voices, and the voice to be identified can be used as the target voice. Then, the sample voices and target voices are processed through subsequent steps S102-103 to determine the word information corresponding to the same syllables to be recommended. This information is then used for subsequent voiceprint identification operations such as deviation analysis, stability analysis, and prosodic feature analysis, thereby realizing the voiceprint comparison and identification of the target user.
[0079] It should be noted that both the sample speech and the target speech can be obtained through recording or other means as needed. This application does not limit the method of obtaining the sample speech and the target speech to be identified.
[0080] S102: Using a pre-built homonym database, perform syllable recognition on the sample speech and the target speech to obtain the words contained in the sample speech and the word frequency corresponding to each word.
[0081] In this embodiment, after obtaining the sample speech and the target speech to be identified through step S101, in order to quickly and accurately recommend the word information corresponding to the same Chinese pronunciation, thereby effectively improving the efficiency and convenience of voiceprint identification, a homophone database is constructed in advance using stop words, neutral tone words, experience words, etc. identified in the previous daily identification process. Then, the sample speech and the target speech are syllable identified separately using the pre-constructed homophone database to obtain the words contained in the sample speech and the target speech and the word frequency corresponding to each word, which is then used to execute the subsequent step 102.
[0082] A syllable is a phonetic unit composed of phonemes (including consonants and vowels). In pinyin languages, syllables are composed of consonants and vowels, while in Chinese, syllables are composed of initials and finals. Generally speaking, one pinyin character constitutes one syllable, for example: "ni1, wo3, sheng4". The number at the end of the syllable indicates the tone: 1 for first tone, 2 for second tone, 3 for third tone, and 4 for fourth tone.
[0083] A phoneme is the smallest unit of speech determined by the natural properties of sound. It is analyzed based on the articulation actions within a syllable; one action constitutes one phoneme. Phonemes are divided into two main categories: vowels and consonants. There are 32 vowels and consonants in Mandarin Chinese. For ease of representation, this application uses initials, simple vowels, and compound vowels as basic phonemes, as shown in Table 1 below:
[0084]
[0085] Table 1
[0086] It should be noted that the homonym database may include, but is not limited to, at least one of the stop word database, the neutral tone word database, and the empirical homonym database. Based on this, a preferred implementation is to use the pre-built stop word database, the neutral tone word database, and the empirical homonym database to jointly perform syllable recognition on the sample speech and the target speech, so as to more accurately identify the words contained in the sample speech and the target speech and the word frequency corresponding to each word.
[0087] Specifically, everyday voice conversation files contain a large number of stop words and unstressed words. To improve the accuracy of identification, this application typically prioritizes the analysis and processing of non-stop words and non-unstressed words during voice identification. Therefore, to distinguish between stop words and unstressed words, this application can pre-construct a stop word lexicon and an unstressed word lexicon.
[0088] Stop words can be words without any real meaning, used to help describe nouns and express concepts in the text without any semantic meaning related to the test questions, such as locations or quantities. The stop word lexicon is arranged alphabetically in descending order of the frequency of stop words, and denoted as S. An example of a stop word lexicon is shown in Table 2 below:
[0089]
[0090]
[0091] Table 2
[0092] In speech sequences, many syllables often lose their original tone and are pronounced with a light and short tone; such words can be classified as unstressed words. Similar to stop word dictionaries, unstressed word dictionaries can also be arranged alphabetically in descending order of frequency of unstressed words, denoted as L. An example of an unstressed word dictionary is shown in Table 3 below:
[0093]
[0094]
[0095] Table 3
[0096] In addition, during routine identification processes, identification personnel also summarize many empirical homophones based on the identification scenarios. In order to improve the identification accuracy, this application has also pre-constructed an empirical homophone lexicon, denoted as J. The empirical homophone lexicon is shown in Table 4 below:
[0097] Experienced homonym dictionary A: Security B: Beijing background collapse C: Financial and Property Dispatch D: Be bold, everyone should be careless, big or small. E: And Tencent F: Selling ……
[0098] Table 4
[0099] It should be noted that there is overlap between the stop word database and the neutral tone word database. However, once a word is marked as an empirical homonym, the corresponding entry will be deleted from both the stop word database and the neutral tone word database. That is, the empirical homonym database does not overlap with the stop word database and the neutral tone word database.
[0100] In one possible implementation of this embodiment, such as Figure 2 As shown, the specific implementation process of using a pre-built stop word lexicon, neutral tone word lexicon, and empirical homophone word lexicon to perform syllable recognition on sample speech and target speech, and obtaining the words contained in the sample speech and target speech and the word frequencies corresponding to each word, can include the following steps (1)-(4):
[0101] Step (1): Perform speech recognition on the sample speech and the target speech respectively to obtain the sample syllable sequence and sample Chinese character document contained in the sample speech, and the target syllable sequence and target Chinese character document contained in the target speech.
[0102] In this implementation, after obtaining the sample speech and the target speech to be identified, in order to quickly and accurately recommend the word information corresponding to the same Chinese pronunciation, thereby effectively improving the efficiency and convenience of voiceprint identification, existing or future speech recognition methods can be used to perform speech recognition on the sample speech and the target speech respectively, so as to identify the syllable sequence and Chinese character document contained in the sample speech, and define them as the sample syllable sequence and sample Chinese character document respectively, and identify the syllable sequence and Chinese character document contained in the target speech, and define them as the target syllable sequence and target Chinese character document respectively.
[0103] Examples of the sample syllable sequence and the target syllable sequence can be shown in Table 5:
[0104] Pinyin name with tone Corresponding Chinese character name Start time End time
[0105] Table 5
[0106] Among them, "Pinyin name with tone" refers to a syllable in the sample syllable or the target syllable; "Corresponding Chinese character name" refers to the name of the entry corresponding to the syllable; "Start time" and "End time" refer to the start time and end time of the syllable in the sample speech or the target speech, respectively.
[0107] Step (2): Use the sample syllable sequence to segment the sample Chinese document to obtain the sample preliminary word entries corresponding to the sample Chinese document; and use the target syllable sequence to segment the target Chinese document to obtain the target preliminary word entries corresponding to the target Chinese document.
[0108] In this implementation, after obtaining the sample Chinese character document corresponding to the sample speech and the target Chinese character document corresponding to the target speech through step (1), the sample Chinese character document and the target Chinese character document can be further processed by word segmentation to obtain the preliminary sample terms corresponding to the sample Chinese character document and the preliminary target terms corresponding to the target Chinese character document, which are then used to execute the subsequent step (3). The specific composition structure of the preliminary sample terms and the preliminary target terms is as follows: Figure 3 As shown, it includes the term name (name), the pinyin name with tones (pinyin), and the number of occurrences (num). Based on the number of occurrences (num), it displays the start time (the specific unit is not limited, but it can usually be in seconds) and end time (the specific unit is not limited, but it can usually be in seconds) of each occurrence of the term's corresponding speech in the sample or target speech, along with the corresponding index (index). For example, when the initial term in the sample or target is "Financial Center" and has appeared only once, its specific structure can be illustrated as follows: Figure 4 As shown.
[0109] Specifically, we first define the sample Chinese character document as D, and then use existing or future word segmentation methods to segment it, which yields the preliminary sample terms contained in the sample Chinese character document D, which are defined as W. The set of the preliminary sample terms W can be represented as follows:
[0110] W = {W|W1,W2,W3…W} M ,W∈wordCut(D)}
[0111] Among them, wordCut() represents the function of word segmentation processing of a document; W i This represents the word i extracted after segmenting the sample Chinese document D.
[0112] It is understandable that the word segmentation process for the target Chinese document can be carried out by referring to the word segmentation process for the sample Chinese document D described above. The specific implementation process will not be repeated here.
[0113] Step (3): Merge identical terms in the preliminary sample terms to obtain merged preliminary sample terms; and merge identical terms in the preliminary target terms to obtain merged preliminary target terms.
[0114] In this implementation, after obtaining the sample preliminary terms corresponding to the sample Chinese character document and the target preliminary terms corresponding to the target Chinese character document through step (2), the sample preliminary terms can be further merged with the same terms to obtain the merged sample preliminary terms; and the target preliminary terms can also be merged with the same terms to obtain the merged target preliminary terms, which are used to execute the subsequent step (4).
[0115] Specifically, one possible implementation is to take the preliminary sample terms corresponding to the sample Chinese document as an example. This involves merging identical terms from all the obtained preliminary sample terms, keeping only one identical term, and incrementing the occurrence count by 1 for each merged term. This results in a merged preliminary sample term set, which is defined as C. After sorting the merged terms in ascending alphabetical order, the resulting set of preliminary sample terms C can be represented as follows:
[0116] C = {C|C1,C2,C3…C} m}
[0117] Among them, C i This represents the initial term i of the merged sample; and C i The alphabetical order of the entries is less than C. i+1 The order of the entries.
[0118] For example, if the initial term in the merged sample is "financial center" and its corresponding audio appears three times in the sample speech, then an example of its specific structural composition can be shown as follows: Figure 5 As shown.
[0119] Understandably, the merging process for identical terms in the initial target terms can be similar to the merging process for identical terms in the initial sample terms described above. The specific implementation process will not be elaborated here. Let the merged initial target terms be defined as B. After sorting them in ascending alphabetical order, the set of merged initial target terms B can be represented as follows:
[0120] B = {B|B1,B2,B3…B} n}
[0121] Among them, B i Indicates the initial target term i after the merger; and B i The alphabetical order of the entry is less than B. i+1 The order of the entries.
[0122] Step (4): Using the pre-built stop word lexicon, neutral tone word lexicon and empirical homophone lexicon, calculate the word frequency of the preliminary sample words after merging, and calculate the word frequency of the preliminary target words after merging.
[0123] In this implementation, after obtaining the merged preliminary sample terms and the merged preliminary target terms through step (3), the word frequency corresponding to the merged preliminary sample terms and the word frequency corresponding to the merged preliminary target terms can be calculated by using the pre-constructed stop word lexicon, neutral tone word lexicon and empirical homophone lexicon as shown in Tables 2, 3 and 4 above, so as to execute the subsequent step S103.
[0124] One possible implementation is to first use a pre-built stop word lexicon, unstressed word lexicon, and empirical homophone lexicon to determine whether the merged preliminary sample entries belong to the stop word, unstressed word, or empirical homophone lexicon. In other words, it is to determine whether the merged preliminary sample entries are stop words, unstressed words, or empirical homophones. If so, the word frequency of the merged preliminary sample entries is calculated using the preset recommendation weights corresponding to the stop words, unstressed words, or empirical homophones. Similarly, the word frequency of the merged target preliminary entries can be calculated.
[0125] The recommended weights corresponding to the preset stop words (S), neutral tone words (L), and empirical homophones (J) can be defined as x, y, and z, respectively. The specific values of the three can be set according to the actual situation and empirical values. This application does not limit this. For example, the recommended weights x, y, and z corresponding to the stop words (S), neutral tone words (L), and empirical homophones (J) can be set to 0.95, 0.8, and 1.5, respectively.
[0126] Specifically, if the initial term C of the merged sample i If it belongs to a stop word, then C is calculated using the preset recommendation weight x corresponding to the stop word. i The corresponding formula for word frequency is as follows:
[0127]
[0128] Among them, F i This indicates the initial term C of the merged sample. i word frequency; C i [num] represents C i The maximum number of occurrences of a term in set C is represented by Max(C[num]), i.e., Max(C[num]) = max({C1[num],C2[num],C3[num]…C…). m [num]}.
[0129] If the initial term C of the merged sample i If it belongs to the category of words with a neutral tone, then C is calculated using the preset recommendation weight y corresponding to words with a neutral tone. i The corresponding formula for word frequency is as follows:
[0130]
[0131] Among them, F i This indicates the initial term C of the merged sample. i word frequency; C i [num] represents C iThe maximum number of occurrences of a term in set C is represented by Max(C[num]), i.e., Max(C[num]) = max({C1[num],C2[num],C3[num]…C…). m [num]}.
[0132] It should be noted that if the initial term C of the merged sample... i If a word is both a stop word and a word with a neutral tone, then one of them can be chosen for word frequency calculation based on the actual situation. A preferred approach is to use the preset recommendation weight x corresponding to the stop word to calculate C. i Corresponding word frequency.
[0133] If the initial term C of the merged sample i If it belongs to the category of empirical homophones, then C is calculated using the preset recommendation weight z corresponding to empirical homophones. i The corresponding formula for word frequency is as follows:
[0134]
[0135] Among them, F i This indicates the initial term C of the merged sample. i word frequency; C i [num] represents C i The maximum number of occurrences of a term in set C is represented by Max(C[num]), i.e., Max(C[num]) = max({C1[num],C2[num],C3[num]…C…). m [num]}; Let represent the average number of occurrences of each term in set C, i.e.
[0136] If the initial term C of the merged sample i If a word is neither a stop word, nor a word with a neutral tone, nor a word with empirical homonyms, then when calculating the word frequency corresponding to Ci using the above formula, z should be set to 1.
[0137] Thus, the set of word frequencies corresponding to the initial terms in the merged sample can be represented as follows:
[0138] F = {F|F1,F2,F3…F} m}
[0139] Among them, F i This represents the initial term C of the merged samples in set C. i Word frequency.
[0140] Understandably, the calculation method for the word frequency of the merged target preliminary terms can be the same as the calculation process for the word frequency of the merged sample preliminary terms described above. The specific calculation process will not be repeated here. The merged target preliminary terms B... i The word frequency is defined as E. After sorting the words in ascending alphabetical order, the resulting set of word frequencies corresponding to the initial target terms can be represented as:
[0141] E = {E|E1,E2,E3…E} n}
[0142] Among them, E i This represents the initial target term B after merging in set B. i Word frequency.
[0143] S103: Construct a word matrix and triples using the words contained in the sample speech and the target speech, as well as the word frequencies corresponding to each word.
[0144] In this embodiment, after obtaining the terms contained in the sample speech and the target speech and the corresponding word frequencies of each term through step S102, in order to quickly and accurately recommend term information corresponding to the same Chinese pronunciation, and effectively improve the efficiency and convenience of voiceprint identification, the terms contained in the sample speech and the target speech and the corresponding word frequencies of each term can be used to construct a term matrix and triples, which are then used to execute the subsequent step S104 to determine the term information to be recommended.
[0145] Specifically, one possible implementation is to first construct a word matrix by using the words contained in the sample speech as rows and the words contained in the target speech as columns. Then, iterate through the matrix and compare whether the words contained in the sample speech in the rows and the words contained in the target speech in the columns are the same. If they are the same, multiply the frequency of the same words in the sample speech by the frequency of the same words in the target speech, and use the product to construct a triplet.
[0146] In this implementation, a matrix with m rows and n columns can be constructed first. The initial sample words C after merging are used as the values of the matrix rows, and the initial target words B after merging are used as the values of the matrix columns to construct a word matrix. This matrix is used to store the product of the word frequencies of the same words in the sample speech and the target speech, denoted as matrix[n][m]. The initial values of all words can be set to 0. The triple structure constructed later is denoted as triple(row, column, value).
[0147] Then, iterate through and compare the terms in C and B (i.e., the rows and columns of the term matrix) to see if they are the same. If term C... i Equal to entry Bj That is, C[i][name] = B[j][name], then matrix[i][j] = F[i] * E[j] * 1000, where multiplying by 1000 is to enlarge the value for easier user observation; F[i] represents the term C. i The word frequency, E[j] represents word B. j The word frequency. And the positions of the non-zero elements in the word matrix i,j (i.e., C) are determined. i Entry and B j The term C and its value matrix[i][j] are stored in the triple. Otherwise, if term C... i Not equal to entry B j If C[i][name]! = B[j][name], then matrix[n][m] = 0.
[0148] For example, an example of the final constructed term matrix can be shown in Table 6 below, and an example of the corresponding triples can be shown in Table 7 below:
[0149]
[0150] Table 6
[0151] row column value 1 3 6 3 2 9 4 m 4
[0152] Table 7
[0153] As shown in Table 6, term B1 equals term C3. Therefore, the product of the corresponding term frequencies, 6, can be used as the value of the triple when row=1 and column=3, as shown in Table 7. Similarly, the triples shown in Table 7 can be constructed based on the term matrix in Table 6.
[0154] S104: Sort the triples and, based on the processing results, determine the word information corresponding to the same syllable to be recommended, for voiceprint identification.
[0155] In this embodiment, after constructing the triplet in step S103, in order to quickly and accurately recommend word information corresponding to the same Chinese pronunciation, thereby effectively improving the efficiency and convenience of voiceprint identification, the triplet can be further sorted according to a preset sorting method (this application does not limit the specific sorting method), and based on the processing result, the word information corresponding to the same syllable to be recommended is determined and the same pronunciation is recommended, such as... Figure 2 As shown, the operational steps for subsequent voiceprint identification, such as deviation analysis, stability analysis, and prosodic feature analysis, are used to improve identification efficiency and convenience.
[0156] Specifically, one possible implementation is to first sort the triples in descending order according to the product of word frequencies (i.e., the values of value in Table 7), and obtain the sorting result. For example, Table 7 can be sorted in descending order according to the value, and the sorting result is shown in Table 8 below:
[0157] row column value 3 2 9 1 3 6 4 m 4
[0158] Table 8
[0159] Then, the identical words corresponding to the product of the first N word frequencies in the sorted results (i.e., the values of 'value' in Table 8) can be identified as the words corresponding to the identical syllables to be recommended. Here, N is a positive integer greater than 0, and N represents the preset total number of recommended words corresponding to the same Chinese pronunciation. Its value can be predefined according to actual circumstances; this application does not impose any limitations on it. For example, if N is set to 10, then the identical words corresponding to the product of the first 10 word frequencies in the sorted results can be used as the words corresponding to the identical syllables to be recommended. Alternatively, if the value of N is not defined in actual application, then the identical words corresponding to the product of the word frequencies of all elements in the triplet can be taken as the words corresponding to the identical syllables to be recommended.
[0160] Next, the start time, end time, and word frequency of the words corresponding to the same syllables to be recommended can be determined in the sample speech and target speech. The start time, end time, and word frequency are then used to construct the word information corresponding to the same syllables to be recommended, which is then used for voiceprint identification.
[0161] Specifically, recommended homophonic terms and their corresponding time slice indices can be constructed based on the rows and columns in the sorted triples (as shown in Table 8). Here, row corresponds to the merged preliminary sample term C in the sample speech, and column corresponds to the merged preliminary sample term B in the target speech. Finally, the term information corresponding to N homophonic syllables is determined and homophonic recommendations are made, including the term name, the pinyin name, the term frequency (wf), the start time (s) and end time (s) of each occurrence of the term's corresponding speech in the sample and target speech based on the number of occurrences (num), and the corresponding index. For example, if a recommended term appears twice in the sample speech and three times in the target speech, an example diagram of the term information corresponding to the homophonic syllables to be recommended in the sample speech is shown below. Figure 6 As shown in the figure, an example diagram is shown of the proposed word information corresponding to the same syllable to be recommended in the target speech. Figure 7 As shown.
[0162] In summary, the Chinese homophony recommendation method provided in this embodiment first acquires sample speech and target speech to be identified. Then, using a pre-constructed homophony dictionary, syllable recognition is performed on the sample and target speech to obtain the words contained in each speech and their corresponding word frequencies. Next, using the words contained in the sample and target speech and their corresponding word frequencies, a word matrix and triples are constructed. The triples are then sorted, and based on the processing results, the word information corresponding to the homophonous syllables to be recommended is determined for subsequent voiceprint identification. It is evident that this application automatically determines the word information corresponding to the homophonous syllables to be recommended after syllable recognition of the sample and target speech using a pre-constructed homophony dictionary. This allows for faster and more accurate voiceprint comparison and identification of the user to whom the target speech belongs, thereby effectively improving the efficiency and convenience of voiceprint identification.
[0163] Second Embodiment
[0164] This embodiment will introduce a Chinese homophone recommendation device; please refer to the above method embodiment for related content.
[0165] See Figure 8 This is a schematic diagram of the composition of a Chinese homophone recommendation device provided in this embodiment. The device 800 includes:
[0166] The acquisition unit 801 is used to acquire sample speech and target speech to be identified;
[0167] The recognition unit 802 is used to perform syllable recognition on the sample speech and the target speech using a pre-built homophone database, so as to obtain the words contained in the sample speech and the target speech and the word frequency corresponding to each word;
[0168] The construction unit 803 is used to construct a word matrix and triples by utilizing the words contained in the sample speech and the target speech and the word frequency corresponding to each word.
[0169] The determining unit 804 is used to sort the triples and, based on the processing results, determine the word information corresponding to the same syllable to be recommended, for voiceprint identification.
[0170] In one implementation of this embodiment, the identification unit 802 is specifically used for:
[0171] Using a pre-built stop word lexicon, neutral tone word lexicon, and empirical homophone word lexicon, syllable recognition is performed on the sample speech and the target speech to obtain the words contained in the sample speech and the target speech and the word frequency corresponding to each word.
[0172] In one implementation of this embodiment, the identification unit 802 includes:
[0173] The recognition subunit is used to perform speech recognition on the sample speech and the target speech respectively, to obtain the sample syllable sequence and sample Chinese character document contained in the sample speech, and the target syllable sequence and target Chinese character document contained in the target speech;
[0174] The word segmentation subunit is used to segment the sample Chinese character document using the sample syllable sequence to obtain preliminary sample terms corresponding to the sample Chinese character document; and to segment the target Chinese character document using the target syllable sequence to obtain preliminary target terms corresponding to the target Chinese character document.
[0175] The merging subunit is used to perform the same term merging process on the sample preliminary terms to obtain the merged sample preliminary terms; and to perform the same term merging process on the target preliminary terms to obtain the merged target preliminary terms;
[0176] The first calculation subunit is used to calculate the word frequency corresponding to the preliminary sample word entries after merging, and to calculate the word frequency corresponding to the preliminary target word entries after merging, using a pre-built stop word lexicon, a neutral tone word lexicon, and an empirical homonym lexicon.
[0177] In one implementation of this embodiment, the merging subunit is specifically used for:
[0178] The sample preliminary terms are merged with identical terms, and the processed sample preliminary terms are sorted in ascending alphabetical order to obtain merged sample preliminary terms; and the target preliminary terms are merged with identical terms, and the processed target preliminary terms are sorted in ascending alphabetical order to obtain merged target preliminary terms.
[0179] In one implementation of this embodiment, the first computing subunit includes:
[0180] The judgment subunit is used to determine whether the merged sample preliminary entries belong to the stop word lexicon, the neutral tone word lexicon, or the empirical homonym lexicon by using a pre-built stop word lexicon, neutral tone word lexicon, and empirical homonym lexicon.
[0181] The second calculation subunit is used to calculate the word frequency corresponding to the preliminary term of the merged sample if it is determined that the preliminary term of the merged sample belongs to the stop word, neutral tone word or empirical homonym word library.
[0182] In one implementation of this embodiment, the construction unit 803 includes:
[0183] Construct sub-units to use the words contained in the sample speech as rows and the words contained in the target speech as columns to construct a word matrix;
[0184] The comparison subunit is used to sequentially traverse and compare whether the words contained in the sample speech in the rows of the word matrix and the words contained in the target speech in the columns are the same;
[0185] The multiplication subunit is used to multiply the frequency of the same word in the sample speech with the frequency of the same word in the target speech if the condition is met, and to construct a triple using the product result.
[0186] In one implementation of this embodiment, the determining unit 804 includes:
[0187] The arrangement subunit is used to arrange the triples in descending order according to the product of word frequencies to obtain the arrangement result;
[0188] The subunit is used to determine the word corresponding to the same syllable to be recommended by multiplying the product of the N word frequencies in the sorting results; where N is a positive integer greater than 0.
[0189] The subunit is used to determine the start time, end time, and word frequency of the word corresponding to the same syllable to be recommended in the sample speech and the target speech, and to use the start time, end time, and word frequency to construct the word information corresponding to the same syllable to be recommended for voiceprint identification.
[0190] Furthermore, embodiments of this application also provide a Chinese homophone recommendation device, including: a processor, a memory, and a system bus;
[0191] The processor and the memory are connected via the system bus;
[0192] The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform any of the above-described methods for recommending Chinese homophones.
[0193] Furthermore, embodiments of this application also provide a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform any of the above-described methods for recommending Chinese homophones.
[0194] Furthermore, this application also provides a computer program product, which, when run on a terminal device, causes the terminal device to execute any of the above-described methods for recommending Chinese homophones.
[0195] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.
[0196] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0197] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0198] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for recommending homophones in Chinese, characterized in that, include: Acquire sample speech and target speech to be identified; Speech recognition is performed on the sample speech and the target speech respectively to obtain the sample syllable sequence and sample Chinese character document contained in the sample speech, and the target syllable sequence and target Chinese character document contained in the target speech; Using the sample syllable sequence, the sample Chinese character document is segmented to obtain preliminary sample terms corresponding to the sample Chinese character document; and using the target syllable sequence, the target Chinese character document is segmented to obtain preliminary target terms corresponding to the target Chinese character document. The sample preliminary terms are merged to obtain merged sample preliminary terms; and the target preliminary terms are merged to obtain merged target preliminary terms. Using a pre-built stop word lexicon, neutral tone word lexicon, and empirical homophone lexicon, as well as preset recommendation weights for stop words, neutral tone words, or empirical homophones, the word frequency of the preliminary term in the merged sample is calculated, and the word frequency of the preliminary term in the merged target is also calculated. Construct a word matrix by using the words contained in the sample speech as rows and the words contained in the target speech as columns; Iterate through the word matrix sequentially and compare whether the words contained in the sample speech in the rows and the words contained in the target speech in the columns are the same; If so, multiply the word frequency of the same word in the sample speech with the word frequency of the same word in the target speech, and use the product result to construct a triple; The triples are sorted in descending order according to the product of word frequencies to obtain the sorting result. Based on the sorting result, the word information corresponding to the same syllable to be recommended is determined for voiceprint identification.
2. The method according to claim 1, characterized in that, The process of merging identical terms in the preliminary sample terms to obtain merged preliminary sample terms, and merging identical terms in the preliminary target terms to obtain merged preliminary target terms, includes: The sample preliminary terms are merged with identical terms, and the processed sample preliminary terms are sorted in ascending alphabetical order to obtain merged sample preliminary terms; and the target preliminary terms are merged with identical terms, and the processed target preliminary terms are sorted in ascending alphabetical order to obtain merged target preliminary terms.
3. The method according to claim 1, characterized in that, The step of calculating the word frequency of the preliminary entries in the merged sample by utilizing a pre-built stop word lexicon, a neutral tone word lexicon, and an empirical homophone lexicon, as well as preset recommendation weights corresponding to stop words, neutral tone words, or empirical homophones, includes: Using a pre-built stop word lexicon, neutral tone word lexicon, and empirical homonym lexicon, determine whether the merged sample preliminary entries belong to the stop word lexicon, neutral tone word lexicon, or empirical homonym lexicon; If so, the word frequency corresponding to the preliminary word entries of the merged sample is calculated by using the preset recommendation weights corresponding to stop words, unstressed words, or empirical homophones.
4. The method according to claim 1, characterized in that, The step of determining the word information corresponding to the same syllable to be recommended based on the arrangement result, for use in voiceprint identification, includes: The products of the first N word frequencies in the sorting results correspond to the same words, which are then used to determine the words corresponding to the same syllables to be recommended; where N is a positive integer greater than 0. The start time, end time, and word frequency of the word corresponding to the same syllable to be recommended are determined in the sample speech and the target speech. The start time, end time, and word frequency are used to construct the word information corresponding to the same syllable to be recommended, which is then used for voiceprint identification.
5. A Chinese homophone recommendation device, characterized in that, include: The acquisition unit is used to acquire sample speech and target speech to be identified; The recognition unit is used to perform syllable recognition on the sample speech and the target speech using a pre-built homophone database, so as to obtain the words contained in the sample speech and the target speech and the word frequency corresponding to each word; The identification unit includes: The recognition subunit is used to perform speech recognition on the sample speech and the target speech respectively, to obtain the sample syllable sequence and sample Chinese character document contained in the sample speech, and the target syllable sequence and target Chinese character document contained in the target speech; The word segmentation subunit is used to segment the sample Chinese character document using the sample syllable sequence to obtain preliminary sample terms corresponding to the sample Chinese character document; and to segment the target Chinese character document using the target syllable sequence to obtain preliminary target terms corresponding to the target Chinese character document. The merging subunit is used to perform the same term merging process on the sample preliminary terms to obtain the merged sample preliminary terms; and to perform the same term merging process on the target preliminary terms to obtain the merged target preliminary terms; The first calculation subunit is used to calculate the word frequency of the preliminary sample entries after merging, and to calculate the word frequency of the preliminary target entries after merging, using a pre-built stop word lexicon, a neutral tone word lexicon, and an empirical homophone lexicon, as well as the preset recommendation weights corresponding to stop words, neutral tone words, or empirical homophones. The construction unit is used to construct a word matrix and triples by utilizing the words contained in the sample speech and the target speech and the word frequency corresponding to each word. The building unit includes: Construct sub-units to use the words contained in the sample speech as rows and the words contained in the target speech as columns to construct a word matrix; The comparison subunit is used to sequentially traverse and compare whether the words contained in the sample speech in the rows of the word matrix and the words contained in the target speech in the columns are the same; The multiplication subunit is used to multiply the word frequency of the same word in the sample speech with the word frequency of the same word in the target speech if the product is true, and to construct a triple using the product result. The determining unit is used to sort the triples in descending order according to the product of word frequencies to obtain the sorting result, and to determine the word information corresponding to the same syllable to be recommended based on the sorting result, so as to perform voiceprint identification.
6. A Chinese pronunciation recommendation device, characterized in that, include: Processor, memory, system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1-4.
7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the method described in any one of claims 1-4.