Voiceprint recognition method and system based on automatic voiceprint identification

Abstract

The present invention relates to the field of artificial intelligence technology, and specifically to a voiceprint recognition method and system based on automatic voiceprint identification. The method comprises: identifying the corresponding voiceprint features of the acquired speech to be identified based on a voiceprint recognition model; comparing the similarity between the identified voiceprint features and the voiceprint library to obtain similar speech, and scoring the similarity of the voiceprint features; extracting the stability features and difference features of the recorded common word voice segments through automatic voiceprint identification and performing automatic voiceprint identification scoring; weighting the voiceprint recognition score and the automatic voiceprint identification score, and outputting the final score. The present invention effectively improves the recall rate and precision rate of voiceprint recognition by means of automatic voiceprint identification technology; calculates the value of the stability features and difference features of different people, and effectively avoids the problem of missed recognition due to large differences in similarity of the same person, and the problem of misidentification of different people due to similarity.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a voiceprint recognition method and system for improving the recall rate and precision rate of voiceprint recognition based on automatic voiceprint identification. Background Art

[0002] With the widespread application of voiceprint recognition technology in public security, financial anti-fraud, criminal investigation and other fields, its application scenarios are increasing and the requirements for related technical indicators are becoming higher and higher. At present, voiceprint recognition technology has made great progress, but the recall rate and accuracy of voiceprint verification still have a lot of room for improvement. Current voiceprint recognition relies more on the expressive power of the model and some back-end processing technologies, which mainly rely on the voiceprint features extracted by the model. However, relying more on the expressive power of the model will have certain shortcomings:

[0003] First, the voiceprint recognition result is the similarity between voiceprint features. However, the voiceprint features extracted from the same person at different times may also have significant differences, resulting in large differences in the similarity of the voiceprints of the same person and a lower ranking.

[0004] Secondly, human voiceprints have stability characteristics and differentiation characteristics. How to effectively use the stability characteristics to distinguish different people and use the differentiation characteristics to determine whether they are the same person, the expression ability of the existing voiceprint recognition model needs to be improved.

[0005] Finally, some existing back-end processing technologies can improve the recognition ability of voiceprint models to a certain extent, but the most basic ones are still based on the voiceprint features extracted by the voiceprint model. Summary of the Invention

[0006] To address the above issues, the present invention aims to provide a voiceprint recognition method based on automatic voiceprint identification technology to improve the recall rate and precision of voiceprint recognition. Automatic voiceprint identification technology based on voiceprint recognition technology can effectively avoid missed recognition due to large differences in similarity between the same person, as well as misidentification due to similarities between different people.

[0007] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0008] In a first aspect, in one embodiment provided by the present invention, a voiceprint recognition method based on automatic voiceprint identification is provided, comprising the following steps:

[0009] Perform voiceprint recognition on the acquired speech to be recognized based on the voiceprint recognition model, and output the corresponding voiceprint features;

[0010] Traverse the voiceprint database, compare the similarity with the output voiceprint features, obtain a sorted list of similar voices, and score the similarity of the voiceprint features to obtain the voiceprint recognition score;

[0011] Performing voice phoneme recognition on the acquired voice to be recognized and the similar voices in the obtained similar voice ranking list through automatic voiceprint identification, and recording the common word voice segments;

[0012] Traversing the speech phonemes, searching for some speech segments with a "consonant-vowel-consonant" structure, using the stable features in the "consonant-vowel-consonant" structure of the speech segments to identify different people, and using the unstable features in the "consonant-vowel-consonant" structure of the speech segments to identify whether they are the same person, searching for some speech segments with the same "consonant-vowel-consonant" structure, extracting stability features and difference features from the recorded common words through automatic voiceprint identification, and scoring the similarity of the stability features and the difference features to obtain an automatic voiceprint identification score;

[0013] The voiceprint recognition score and the automatic voiceprint identification score are weighted and summed to output a final score.

[0014] In some embodiments provided by the present invention, the voiceprint library stores the speech to be compared, and when the output voiceprint features are compared with the voiceprint features in the voiceprint library for similarity, the PLDA or cosine scoring method is used for comparison.

[0015] In some embodiments provided by the present invention, the top 20 similar voices in the similar voice ranking list are selected for automatic voiceprint identification.

[0016] In some embodiments provided by the present invention, the automatic voiceprint identification is to use the experience and rules of identification technology to perform automatic comparison and analysis based on the recommended range given by automatic voiceprint recognition to find key evidence. The method of the automatic voiceprint identification is:

[0017] Preprocessing the acquired speech to be recognized;

[0018] Perform feature extraction on the pre-processed speech to be recognized;

[0019] The extracted features are identified by a speech recognition model to obtain the model output results;

[0020] Decode the model output result and mark the phonemes or Chinese characters and their starting and ending positions to obtain the decoding result and the phoneme or Chinese character starting and ending position marks.

[0021] In some embodiments provided by the present invention, Jieba word segmentation or other word segmentation methods and the same string matching algorithm are used to record the common word segments of the speech to be recognized and the top 20 similar speech.

[0022] In some embodiments provided by the present invention, when the automatic voiceprint identification extracts stability features and difference features from a recorded voice segment, when the first formant and the second formant are stability features, and the third formant and the fourth formant are unstable features, the method for extracting stability features and difference features from a recorded voice segment by the automatic voiceprint identification includes:

[0023] Perform framing, windowing and pre-emphasis processing on the recorded voice segments;

[0024] Performing linear prediction coding or spectrum calculation on the processed frame segments to obtain the resonance peaks of the corresponding frames;

[0025] Continuous frame formant extraction is performed on all frames of the recorded speech segment to obtain the first formant, the second formant, the third formant and the fourth formant for feature comparison.

[0026] In some embodiments provided by the present invention, the method for scoring the stability feature similarity and the difference feature similarity is: taking the numerical values of the first formant and the second formant, the third formant and the fourth formant in multiple frames of the recorded speech segment as a vector, performing statistical analysis on the data in the multiple frames and assigning different weights to obtain the final similarity score.

[0027] In some embodiments provided by the present invention, when weighted summing is performed on the voiceprint recognition score and the automatic voiceprint identification score, the greater the number of recorded voice segments, the greater the weight of the automatic voiceprint identification score.

[0028] In some embodiments provided by the present invention, the voiceprint recognition score and the automatic voiceprint identification score are weighted and summed to output a final score. The calculation formula is:

[0029]

[0030] Where, is (1- - - -...- ) ; for + + + ... ;in, 、 、 The value of is related to the speech rate, vowel type, and spectral clarity of the speech to be recognized and similar speech.

[0031] On the second aspect, in another embodiment provided by the present invention, a voiceprint recognition system based on automatic voiceprint identification is provided, and the voiceprint recognition system based on automatic voiceprint identification adopts the aforementioned voiceprint recognition method based on automatic voiceprint identification to calculate the value of stability characteristics and difference characteristics of different people; the voiceprint recognition system based on automatic voiceprint identification includes a voiceprint recognition module, a speech comparison module, an automatic voiceprint identification module, a scoring module and a weight summation calculation module.

[0032] The voiceprint recognition module is used to perform voiceprint recognition on the input speech to be recognized through the voiceprint recognition model and output the corresponding voiceprint features;

[0033] The voice comparison module is used to compare the voiceprint features of the voice to be recognized with the voices in the voiceprint library to obtain similar voices that are close to the same person;

[0034] The automatic voiceprint identification module is used to perform voice phoneme recognition on the acquired voice to be recognized and the similar voices in the obtained similar voice sorting list, and record the shared word voice segments; it is also used to traverse the voice phonemes, and search for some voice segments with a "consonant-vowel-consonant" structure. The stable features in the "consonant-vowel-consonant" structure of the partial voice segments are used to identify different people, and the unstable features in the "consonant-vowel-consonant" structure of the partial voice segments are used to identify whether they are the same person. The partial voice segments with the same "consonant-vowel-consonant" structure are searched, and the stability features and difference features of the recorded shared word voice segments are extracted through automatic voiceprint identification.

[0035] The scoring module is used to score the voiceprint feature similarity output by the voiceprint recognition model to obtain a voiceprint recognition score; it is also used to score the stability feature similarity and the difference feature similarity to obtain an automatic voiceprint identification score;

[0036] The weighted sum calculation module is used to perform weighted summation on the voiceprint recognition score and the automatic voiceprint identification score, and output a final score.

[0037] In some embodiments provided by the present invention, the voiceprint recognition system based on automatic voiceprint identification further includes a word segmentation module, which is used to record word segments shared by the speech to be recognized and similar speech using the stuttering word segmentation method and the same string matching algorithm.

[0038] The technical solution provided by the present invention has the following beneficial effects:

[0039] The voiceprint recognition method and system based on automatic voiceprint identification provided by the present invention effectively improve the recall rate and precision rate of voiceprint recognition by leveraging automatic voiceprint identification technology. The system calculates the value of the stability characteristics and differential characteristics of different individuals, and the final result is determined by relying on both voiceprint recognition and automatic voiceprint identification. The final score is output by weighted summation of the voiceprint recognition score and the automatic voiceprint identification score. By implementing automatic voiceprint identification technology based on voiceprint recognition technology, the present invention effectively avoids missed recognition due to large differences in similarity between the same individual, as well as misidentification due to similarities between different individuals.

[0040] These and other aspects of the present invention will become more readily apparent in the following description of the embodiments. It should be understood that the above general description and the following detailed description are merely exemplary and explanatory and are not intended to limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following briefly introduces the drawings required for the exemplary embodiments or related technical descriptions. The drawings are used to provide a further understanding of the present invention and constitute part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the drawings:

[0042] Figure 1 The present invention provides a flow chart of a voiceprint recognition method based on automatic voiceprint identification.

[0043] Figure 2 A schematic diagram of the recognition process of a voiceprint recognition method based on automatic voiceprint identification provided by an embodiment of the present invention.

[0044] Figure 3 The present invention provides a flowchart of automatic voiceprint identification in a voiceprint recognition method based on automatic voiceprint identification.

[0045] Figure 4 A system block diagram of a voiceprint recognition system based on automatic voiceprint identification provided by an embodiment of the present invention.

[0046] Figure 5 A system block diagram of a voiceprint recognition system based on automatic voiceprint identification is provided in accordance with another embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, nor do they limit "first" and "second" to be different types.

[0049] The following will clearly and completely describe the technical solutions in the exemplary embodiments of the present invention in conjunction with the accompanying drawings of the exemplary embodiments of the present invention. Obviously, the exemplary embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] Today's voiceprint recognition relies more on the expressive power of the model. At different times, the voiceprint features extracted from the same person may also differ significantly, resulting in large differences in the similarity of the same person's voiceprints and a low ranking. Since human voiceprints have stability characteristics and differentiation characteristics, how to effectively use stability characteristics to distinguish different people and use differentiation characteristics to determine whether they are the same person requires improvement in the expressive power of existing voiceprint recognition models. Moreover, some existing back-end processing technologies can improve the recognition ability of voiceprint models to a certain extent, but the most basic of them is still based on the voiceprint features extracted by the voiceprint model.

[0051] To solve the above problems, an embodiment of the present invention provides a voiceprint recognition method based on automatic voiceprint identification technology. Automatic voiceprint identification technology is performed on the basis of voiceprint recognition technology, which can effectively avoid missed identification caused by large differences in similarity of the same person, and misidentification caused by similarities between different people.

[0052] The technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.

[0053] Example 1

[0054] See Figure 1 and Figure 2 As shown, Figure 1 This is a flow chart of a voiceprint recognition method based on automatic voiceprint identification provided by the present invention. Figure 2The present invention provides a voiceprint recognition method based on automatic voiceprint identification, which includes the following steps:

[0055] S1: Perform voiceprint recognition on the acquired speech to be recognized based on the voiceprint recognition model, and output the corresponding voiceprint features.

[0056] In this embodiment, the speech to be recognized is input into the voiceprint recognition model, which then outputs the corresponding voiceprint features. When training the voiceprint recognition model, the training dataset can be a public dataset, an internal company dataset, or a mixture thereof. In this embodiment, public datasets include the voxceleb1 and voxceleb2 datasets, the aidatatang dataset, the LibriSpeech dataset, the aishell dataset, and the CN-Celeb dataset, or historical data samples collected internally by the company.

[0057] In the voiceprint recognition model, the corresponding voiceprint features are mainly embedded in the neural network and a set of vector values after the fully connected layer. The voiceprint recognition model can output the voiceprint features corresponding to the speech to be recognized.

[0058] S2: Traverse the voiceprint database, compare the similarity with the output voiceprint features, obtain a sorted list of similar voices, and score the similarity of the voiceprint features to obtain the voiceprint recognition score.

[0059] It should be noted that the output voiceprint features are compared with the voiceprint features of the speech to be compared in the voiceprint database for similarity. Among them, when the output voiceprint features are compared with the voiceprint features of the speech to be compared in the voiceprint database for similarity, the PLDA or cosine scoring method is used for comparison, which will not be detailed here.

[0060] In this embodiment, when scoring the similarity of voiceprint features, features are extracted from a speech, and then the extracted features are input into the voiceprint recognition model. The voiceprint recognition model performs a series of operations through the model network to output a set of vector values, which can be used as abstract voiceprint features. The similarity of the voiceprint features (i.e., the output vectors) of the two speech sounds is scored to obtain a voiceprint score.

[0061] In this embodiment, a ranked list of similar voices is obtained, and automatic voiceprint identification is performed on the top 20 voices with the highest similarity, and the following steps S3-S5 are performed.

[0062] S3: Perform voice phoneme recognition on the acquired voice to be recognized and the similar voices in the obtained similar voice ranking list through automatic voiceprint identification, and record the common word voice segments.

[0063] In this embodiment, automatic voiceprint identification technology is used to perform voice or phoneme recognition on the target voice and the top 20 voices. A selector is used to select the 20 voices with the highest voiceprint scores, primarily based on the similarity scores of the automatic voiceprint identification. The automatic voiceprint identification is based on the recommended range given by the automatic voiceprint identification and uses the experience and rules of identification technology to automatically compare and analyze to find key evidence. The method of the automatic voiceprint identification is as follows:

[0064] Preprocess the acquired speech to be recognized (VAD, denoising, etc. in speech recognition);

[0065] Perform feature extraction on the pre-processed speech to be recognized;

[0066] The extracted features are identified by a speech recognition model to obtain the model output results;

[0067] Decode the model output result and mark the phonemes or Chinese characters and their starting and ending positions to obtain the decoding result and the phoneme or Chinese character starting and ending position marks.

[0068] In this embodiment, the pre-processing of the speech to be recognized includes processing using a speech recognition endpoint detection (VAD) algorithm and performing speech denoising. When extracting features from the pre-processed speech, the MFCC speech recognition analysis algorithm, the speech PLP feature extraction algorithm, the FBank algorithm, etc. are used for feature extraction.

[0069] It should be noted that, before recording the shared word voice segments, the present invention further includes: using Jieba word segmentation or other word segmentation methods and the same string matching algorithm to record the shared word voice segments between the speech to be recognized and the top 20 similar speech segments.

[0070] S4: Traverse the speech phonemes and search for some speech segments with a "consonant-vowel-consonant" structure. The stable features in the "consonant-vowel-consonant" structure of the speech segments are used to identify different people. The unstable features in the "consonant-vowel-consonant" structure of the speech segments are used to identify whether they are the same person. Search for some speech segments with the same "consonant-vowel-consonant" structure. Extract the stability features and difference features of the recorded common words from the speech segments through automatic voiceprint identification, and score the stability feature similarity and difference feature similarity to obtain the automatic voiceprint identification score.

[0071] In this embodiment, the speech phonemes are traversed, and some of the speech segments searched are of a "consonant-vowel-consonant" structure. The stable features in the "consonant-vowel-consonant" structure of the partial speech segments are used to identify different people, and the unstable features in the "consonant-vowel-consonant" structure of the partial speech segments are used to identify whether they are the same person.

[0072] It should be noted that when the automatic voiceprint identification extracts stability features and difference features from the recorded voice segment, when the first and second formants are stability features, the third and fourth formants are non-stable features, see Figure 3 As shown, the method for extracting stability features and difference features of recorded voice segments by automatic voiceprint identification includes:

[0073] S401, performing framing, windowing and pre-emphasis processing on the recorded voice segment;

[0074] S402, performing linear predictive coding (LPC) or spectrum calculation on the processed frame segment, calculating possible formant of the frame through LPC or spectrum, and obtaining the formant of the corresponding frame;

[0075] S403 , performing continuous frame formant extraction on all frames of the recorded speech segment, obtaining the first formant, the second formant, the third formant, and the fourth formant for feature comparison.

[0076] In this embodiment, the method for scoring the stability feature similarity and the difference feature similarity is: taking the numerical values of the first formant and the second formant, the third formant and the fourth formant in multiple frames of the recorded speech segment as a vector, performing statistical analysis on the data in the multiple frames, calculating statistical indicators such as zero-crossing rate, mean, variance, standard deviation, and finally assigning different weights to these indicators and assigning different weights to obtain the final similarity score.

[0077] S5: performing weighted summation on the voiceprint recognition score and the automatic voiceprint identification score, and outputting a final score.

[0078] It should be noted that when the weighted sum of the voiceprint recognition score and the automatic voiceprint identification score is performed, the greater the number of recorded segments, the greater the weight of the automatic voiceprint identification score.

[0079] Stability feature similarity scoring and difference feature similarity scoring are considered part of automatic voiceprint identification scoring, while voiceprint recognition scoring is considered part of overall voice scoring. Automatic voiceprint recognition scoring provides the top 20 voices that are likely to be from the same person as the voice to be identified. Automatic voiceprint identification scoring specifically analyzes which voices are from the same person as the voice to be identified.

[0080] Perform weighted summation on the voiceprint recognition score and the automatic voiceprint identification score, and output the final score The calculation formula is:

[0081]

[0082] Where, is (1- - - -...- ) ; for + + + ... ;in, 、 、 The value of is related to the speech rate, vowel type, and spectral clarity of the speech to be recognized and similar speech.

[0083] The present invention provides a voiceprint recognition method based on automatic voiceprint identification, which performs automatic voiceprint identification technology on the basis of voiceprint recognition technology, effectively avoiding the problem of missed identification caused by large differences in similarity of the same person, and misidentification caused by similarities between different people.

[0084] It should be understood that, although the above is described in a certain order, these steps are not necessarily performed in sequence according to the above order. Unless there is clear explanation in this article, the execution of these steps does not have strict order restriction, and these steps can be performed in other orders. Moreover, a part of the steps of the present embodiment may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be performed in turn or alternately with at least a portion of the steps or stages in other steps or other steps.

[0085] Example 2

[0086] Figure 4 FIG2 shows a structural block diagram of a voiceprint recognition system based on automatic voiceprint identification provided by Example 2 of the present invention. Figure 4 As shown, in a preferred embodiment of the present invention, a voiceprint recognition system based on automatic voiceprint identification includes a voiceprint recognition module 100, a speech comparison module 200, an automatic voiceprint identification module 300, a scoring module 400 and a weighted sum calculation module 500. Among them:

[0087] The voiceprint recognition module 100 is used to perform voiceprint recognition on the input speech to be recognized through a voiceprint recognition model, and output corresponding voiceprint features.

[0088] In this embodiment, when the voiceprint recognition model is trained, the training data set may be a public data set or an internal company data set, or a mixed data set thereof.

[0089] The corresponding voiceprint features are mainly embedded in the neural network and a set of vector values after the fully connected layer. The voiceprint recognition model can output the voiceprint features corresponding to the speech to be recognized.

[0090] The voice comparison module 200 is used to compare the voiceprint features of the speech to be recognized with the voices in the voiceprint database to obtain similar voices that are approximately from the same person.

[0091] In this embodiment, the output voiceprint features are compared with the voiceprint features of the speech to be compared in the voiceprint database for similarity. PLDA or cosine scoring is used to compare the output voiceprint features with the voiceprint features of the speech to be compared in the voiceprint database for similarity.

[0092] When scoring the similarity of voiceprint features, features are extracted from a speech, and then the extracted features are input into the voiceprint recognition model. The voiceprint recognition model performs a series of operations through the model network and outputs a set of vector values. This set of vector values can be used as an abstract voiceprint feature. The similarity of the voiceprint features (i.e., the output vector) of the two speech sounds is scored to obtain the voiceprint score.

[0093] In this embodiment, a selector is used to select the 20 voices with the highest voiceprint scores, primarily based on the similarity scores of automatic voiceprint recognition. This allows for automatic voiceprint identification technology to perform voice or phoneme recognition on the target voice and the top 20 voices. The resulting ranked list of similar voices is then used for automatic voiceprint identification of the top 20 voices with the highest similarity.

[0094] The automatic voiceprint identification module 300 is used to perform voice phoneme recognition on the acquired voice to be recognized and the similar voice in the obtained similar voice sorting list, and record the common word voice segments; it is also used to traverse the voice phonemes, and search for some voice segments with a "consonant-vowel-consonant" structure. The stable features in the "consonant-vowel-consonant" structure of the some voice segments are used to identify different people, and the unstable features in the "consonant-vowel-consonant" structure of the some voice segments are used to identify whether they are the same person. The partial voice segments with the same "consonant-vowel-consonant" structure are searched for, and the partial voice segments with the same "consonant-vowel-consonant" structure are searched for, and stability features and difference features are extracted.

[0095] When automatic voiceprint identification extracts stability features and difference features from a recorded voice segment, when the first and second formants are stability features and the third and fourth formants are non-stable features, the method for automatic voiceprint identification to extract stability features and difference features from a recorded voice segment includes: framing, windowing, and pre-emphasis processing the recorded voice segment; performing linear predictive coding (LPC) or spectrum calculation on the processed frame segment, calculating possible formants of the frame through LPC or spectrum, and obtaining formants of the corresponding frame; performing continuous frame formant extraction on all frames of the recorded voice segment, obtaining the first and second formants, the third formant, and the fourth formant for feature comparison.

[0096] The scoring module 400 is used to score the voiceprint feature similarity output by the voiceprint recognition model to obtain a voiceprint recognition score; it is also used to score the stability feature similarity and the difference feature similarity to obtain an automatic voiceprint identification score.

[0097] When obtaining the automatic voiceprint identification score, the method for scoring the stability feature similarity and the difference feature similarity is: taking the values of the first formant and the second formant, the third formant and the fourth formant in multiple frames of the recorded voice segment as a vector, performing statistical analysis on the data in the multiple frames, calculating statistical indicators such as zero-crossing rate, mean, variance, standard deviation, and finally assigning different weights to these indicators and assigning different weights to obtain the final similarity score.

[0098] The weighted sum calculation module 500 is used to perform weighted summation on the voiceprint recognition score and the automatic voiceprint identification score, and output a final score.

[0099] In this embodiment, when weighted summing is performed on the voiceprint recognition score and the automatic voiceprint identification score, the greater the number of recorded segments, the greater the weight of the automatic voiceprint identification score.

[0100] Perform weighted summation on the voiceprint recognition score and the automatic voiceprint identification score, and output the final score The calculation formula is:

[0101]

[0102] Where, is (1- - - -...- ) ; for + + + ... ;in, 、 、 The value of is related to the speech rate, vowel type, and spectral clarity of the speech to be recognized and similar speech.

[0103] Figure 5 The following figure shows a structural block diagram of a voiceprint recognition system based on automatic voiceprint identification according to another embodiment of the present invention. The voiceprint recognition system based on automatic voiceprint identification also includes a word segmentation module 600, which is used to record the common words and voice segments between the speech to be recognized and similar speech using the Jieba word segmentation method and the same string matching algorithm.

[0104] In this embodiment, the voiceprint recognition system based on automatic voiceprint identification adopts the steps of the voiceprint recognition method based on automatic voiceprint identification as described above during execution. Therefore, the operation process of the voiceprint recognition system based on automatic voiceprint identification will not be introduced in detail in this embodiment.

[0105] In summary, the technical solution provided by the present invention has the following advantages:

[0106] 1. The voiceprint recognition method and system based on automatic voiceprint identification provided by the present invention effectively improve the recall rate and precision rate of voiceprint recognition with the help of automatic voiceprint identification technology; the value of the stability characteristics and difference characteristics of different people is calculated, and the final result is determined by voiceprint recognition and automatic voiceprint identification. The weighted summation of the voiceprint recognition score and the automatic voiceprint identification score is performed to output the final score.

[0107] 2. The present invention uses automatic voiceprint identification technology based on voiceprint recognition technology to effectively avoid missed recognition caused by large differences in similarity between the same person and misidentification caused by similarities between different people.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A voiceprint recognition method based on automatic voiceprint identification, characterized in that: include: Perform voiceprint recognition on the acquired speech to be recognized based on the voiceprint recognition model, and output the corresponding voiceprint features; Traverse the voiceprint database, compare the similarity with the output voiceprint features, obtain a sorted list of similar voices, and score the similarity of the voiceprint features to obtain the voiceprint recognition score; Performing voice phoneme recognition on the acquired voice to be recognized and the similar voices in the obtained similar voice ranking list through automatic voiceprint identification, and recording the common word voice segments; The speech phonemes are traversed, and some speech segments are searched for that have a "consonant-vowel-consonant" structure. The stable features in the "consonant-vowel-consonant" structure of these speech segments are used to identify different people, and the unstable features in the "consonant-vowel-consonant" structure of these speech segments are used to identify whether they are the same person. Some speech segments with the same "consonant-vowel-consonant" structure are searched, and the stability features and difference features of the recorded common words are extracted through automatic voiceprint identification. The similarity of the stability features and the difference features are scored to obtain the automatic voiceprint identification score; The voiceprint recognition score and the automatic voiceprint identification score are weighted and summed to output a final score.

2. The voiceprint recognition method based on automatic voiceprint identification according to claim 1, characterized in that: The voiceprint database stores the speech to be compared. When the output voiceprint features are compared with the voiceprint features in the voiceprint database for similarity, the PLDA or cosine scoring method is used for comparison.

3. The voiceprint recognition method based on automatic voiceprint identification according to claim 2, characterized in that: The top 20 similar voices in the similar voice ranking list are selected for automatic voiceprint identification.

4. The voiceprint recognition method based on automatic voiceprint identification according to claim 1, characterized in that: The method of automatic voiceprint identification is: Preprocessing the acquired speech to be recognized; Perform feature extraction on the pre-processed speech to be recognized; The extracted features are identified by a speech recognition model to obtain the model output results; Decode the model output result and mark the phonemes or Chinese characters and their starting and ending positions to obtain the decoding result and the phoneme or Chinese character starting and ending position marks.

5. The voiceprint recognition method based on automatic voiceprint identification according to claim 4, characterized in that: When the automatic voiceprint identification extracts stability features and difference features from a recorded voice segment, when the first formant and the second formant are stability features, and the third formant and the fourth formant are unstable features, the method for extracting stability features and difference features from a recorded voice segment by the automatic voiceprint identification includes: Perform framing, windowing and pre-emphasis processing on the recorded voice segments; Performing linear prediction coding or spectrum calculation on the processed frame segments to obtain the resonance peaks of the corresponding frames; Continuous frame formant extraction is performed on all frames of the recorded speech segment to obtain the first formant, the second formant, the third formant and the fourth formant for feature comparison.

6. The voiceprint recognition method based on automatic voiceprint identification according to claim 5, characterized in that: When weighted summing is performed on the voiceprint recognition score and the automatic voiceprint identification score, the more voice segments recorded, the greater the weight of the automatic voiceprint identification score.

7. The voiceprint recognition method based on automatic voiceprint identification according to claim 6, characterized in that: Perform weighted summation on the voiceprint recognition score and the automatic voiceprint identification score, and output the final score The calculation formula is: Where, is (1- - - -...- ) ; for + + + ... ;in, 、 、 The value of is related to the speech rate, vowel type, and spectral clarity of the speech to be recognized and similar speech.

8. A voiceprint recognition system based on automatic voiceprint identification, characterized in that: The voiceprint recognition system based on automatic voiceprint identification uses the voiceprint recognition method based on automatic voiceprint identification according to any one of claims 1 to 7 to calculate the value of stability characteristics and difference characteristics of different people; the voiceprint recognition system based on automatic voiceprint identification includes: The voiceprint recognition module is used to perform voiceprint recognition on the input speech to be recognized through the voiceprint recognition model and output the corresponding voiceprint features; The voice comparison module is used to compare the voiceprint features of the voice to be recognized with the voices in the voiceprint library to obtain similar voices that are close to the same person; The automatic voiceprint identification module is used to perform voice phoneme recognition on the acquired speech to be recognized and similar speech in the obtained similar speech sorting list, and record the shared word voice segments; it is also used to traverse the voice phonemes, searching for some speech segments with a "consonant-vowel-consonant" structure. The stable features in the "consonant-vowel-consonant" structure of the said speech segments are used to distinguish different people, and the unstable features in the "consonant-vowel-consonant" structure of the said speech segments are used to distinguish whether they are the same person. The module searches for some speech segments with the same "consonant-vowel-consonant" structure, and extracts stability features and difference features from the recorded shared word voice segments through automatic voiceprint identification. The scoring module is used to score the similarity of the voiceprint features output by the voiceprint recognition model to obtain the voiceprint recognition score; it is also used to score the similarity of the stability features and the similarity of the difference features to obtain the automatic voiceprint identification score; The weighted sum calculation module is used to perform weighted summation on the voiceprint recognition score and the automatic voiceprint identification score, and output a final score.

9. The voiceprint recognition system based on automatic voiceprint identification according to claim 8, characterized in that: Also includes: The word segmentation module is used to record the word segments shared by the speech to be recognized and similar speech using the Jieba word segmentation method and the same string matching algorithm.

Images