Identity identification method and system based on voiceprint in motion state

By converting the voice sample and samples into broadband speech graphs, selecting formant direction, frequency and transitional phonological analysis methods, the accuracy of the same identification of voiceprints in the motion state is solved, and the identification efficiency and accuracy are improved. In various motion patterns, recording conditions are simplified.

CN118918903APending Publication Date: 2024-11-08吉林警察学院
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410194414.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-02-21
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing method of identifying voiceprints with the same method is difficult to accurately identify voice features in a moving state, resulting in a high misjudgment rate and a wide selection range of featured sound segments, which wastes time or causes misjudgment.

Method used

Using the vocal print identity identification method based on motion state, the resonance peak of characteristic phonemes is analyzed by converting the speech sample and sample into broadband speech graph, and the formant peak direction characteristics, formant frequency characteristics and intrasyllabic transitional signature analysis method are selected to analyze the resonance peak of characteristic phonemes. Direction, frequency and transition characteristics, judge the similarity of voiceprints and provide identification conclusions.

Benefits of technology

It improves the efficiency and accuracy of voiceprint identification, is suitable for various sports forms, without limiting the movement status of the pronunciator when recording, simplifying the recording conditions of voice samples, and improving the identification efficiency and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918903A_ABST
    Figure CN118918903A_ABST
Patent Text Reader

Abstract

The invention discloses a voiceprint identity identification method and system based on a motion state, and the method comprises the steps: firstly obtaining a voice detection material, and carrying out the analysis to obtain the motion state of a speaker; the method comprises the following steps of: acquiring a voice sample, converting the voice sample into a broadband speech graph, selecting N phonemes, performing identity analysis by adopting any one or a combination of more of a formant trend characteristic analysis method, a formant frequency characteristic analysis method and an intra-syllable transition phoneme analysis method after the number of the phonemes meets an identification condition, and outputting a voiceprint identification result. The method is suitable for voiceprint identification in different motion forms, is convenient for online use, and greatly improves the efficiency and accuracy of identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to a method and device for voiceprint identity identification, which belongs to the technical field of voiceprint identification, and in particular to a method and system for identity identification based on voiceprints in motion. Background Art

[0002] Voiceprint identification is a specialized technical means that can provide evidence for trial activities and clues for investigative activities. However, speech and its voiceprint features are dynamic. As a true record of the joint movement of the vocal organs under the control of the brain and nervous system, even if the same person pronounces the same syllable twice and the movement form of the vocal organs remains basically the same, the movement state will not be exactly the same, and then there will be a certain degree of difference in the voiceprint features. However, the actual application environment is more complex and changeable. Affected by many factors such as environmental noise, channel mismatch, disguised pronunciation, and fake intrusion, speech signal processing and speech verification have always been difficult problems in voiceprint identification, which has brought huge challenges to the application of voiceprint identification in judicial practice.

[0003] In judicial practice, due to limited conditions, it is impossible to always ensure that the recording of the inspection material and the sample recording are made by the speaker in the same state. This requires the identification personnel to be familiar with the impact of the speaker's physical state on the voice characteristics and try to use stable voice characteristics for inspection and identification. Before and after exercise, the speaker's body functions are different from usual, and there will be physical changes such as accelerated heartbeat, faster breathing rate, and increased oxygen intake. These changes will affect the movement state of the speaker's vocal organs, thereby changing the voice characteristics.

[0004] At present, the method of voiceprint identification still has the following problems: the selection range of characteristic segments is relatively broad, and it is inevitable that a large number of segments that do not meet the identification requirements will appear, wasting the time of the identification personnel or causing misjudgment by the identification personnel. Summary of the invention

[0005] According to one aspect of the present application, a method for identifying the identity of a voiceprint in motion is provided, which can quickly and accurately complete the voiceprint identity identification, and is particularly suitable for the situation where the speaker is recording voice in motion.

[0006] The method for identifying the identity of a person based on a voiceprint in motion includes:

[0007] Acquiring speech samples, and converting the speech samples and speech samples into broadband speech graphs;

[0008] Select a feature analysis method, select N phonemes, and when N meets the identification conditions, perform feature identity analysis on the broadband speech graph to obtain a voiceprint identification result; wherein the feature analysis method adopts one or more of the formant trend feature analysis method, the formant frequency feature analysis method, and the intra-syllable transition sound feature analysis method.

[0009] Optionally, the method further comprises: after acquiring the voice sample, analyzing the movement state of the speaker; and the feature analysis method is selected according to the movement state of the speaker.

[0010] Optionally, the speaker's motion state includes but is not limited to one or more of still, uniform, accelerating, decelerating, straight, and curved.

[0011] Optionally, the software for converting voice test materials and voice samples into broadband speech graphs may be voice processing software, or various voice workstations, voice analysis systems, and the like.

[0012] Optionally, the formant trend feature analysis method includes: analyzing the curve shape of the formant center line in the broadband spectrum, selecting at least one of the following characteristic phonemes to analyze the formant trend feature: monophthong finals, diphthong finals, and triphthong finals.

[0013] Optionally, the monophthongs include: “a”, “o”, “e”, “i”, “u”, “ü”.

[0014] Preferably, the monophthong finals are "u" and / or "e", and the resonance peak trend characteristics of these two monophthongs at different movement speeds are basically the same and are not affected by the movement speed of the speaker.

[0015] Optionally, the diphthong finals include at least one of the following: “ao”, “ai”, “ia”, “ue”.

[0016] Preferably, the diphthong finals are "ao" and / or "ai". Although the directions of the resonance peaks of these two diphthongs are slightly different at different movement speeds, the overall variation is not large, indicating that the pronunciations of these two diphthongs are very stable and can be used as an important reference for determining or denying identity.

[0017] Optionally, the tri-vowel finals include at least one of the following: “iao”, “iou”, “uai”, “uei”.

[0018] Preferably, the tri-vowel final is "iao", whose resonance peak trends at different movement speeds are basically the same, and the resonance peak trend characteristics at the same movement speed are very stable and are basically not affected by the initial consonant and the following syllable.

[0019] Optionally, the formant frequency characteristic analysis method includes: analyzing the frequency value, mean value, standard deviation and coefficient of variation of the characteristic phoneme formant, and judging whether the coefficient of variation CV1 is within a reasonable range.

[0020] Optionally, the characteristic phonemes include one or more of “a”, “o”, “e”, “i”, “u”, “ü”, “ao”, “ai”, “ia”, “ue”, “iao”, “iou”, “uai”, and “uei”.

[0021] Preferably, the characteristic phonemes are "a", "i", "e" and "ai". By selecting the straight formant in the vowel for frequency measurement, the measurement error is reduced.

[0022] Optionally, the range of the coefficient of variation CV1 is set to be less than 10%, which means that the formant frequency feature analysis item is considered to have passed, otherwise, the next phoneme identification is entered.

[0023] Optionally, the coefficient of variation calculation formula is as follows:

[0024]

[0025]

[0026]

[0027] Where σ is the standard deviation of the formant frequency, x1 and x2 are the formant frequencies of the selected characteristic phonemes in the speech specimen and speech sample, respectively. It is the average value of the formant frequency of the selected characteristic phonemes in the speech specimen and speech sample.

[0028] Preferably, x1 and x2 are average values ​​of the formant frequencies of the selected characteristic phonemes in the speech specimens and speech samples measured multiple times.

[0029] Optionally, the intra-syllable transition phonetic feature analysis method: determines whether the intra-syllable vowel-consonant transition segment morphological features of the characteristic phonemes of the speech specimen and the speech sample are consistent. If they are consistent, the authentication is passed; if not, proceed to the next phoneme identification.

[0030] Optionally, the characteristic phonemes include one or more of “an”, “en”, “in” and “un”.

[0031] Optionally, the characteristic phoneme is "an". Under different movement speeds, the transition sound sign of the phoneme is descending and the transition method is continuous. The vowel-consonant transition sound sign within the syllable is basically not affected by the movement speed, and can be used as an important reference for determining identity and denying identity.

[0032] Optionally, the number of selected phonemes N≥10.

[0033] Preferably, the number of phonemes is 10≤N≤20. Too few phonemes will lead to inaccurate analysis results, and too many phonemes will affect analysis efficiency.

[0034] According to another aspect of the present application, there is provided a system for identifying an identity based on a voiceprint in motion, comprising:

[0035] A data processing module, used for converting speech test materials and speech samples into broadband speech graphs;

[0036] The voiceprint identification module is used to select a feature analysis method, analyze the voiceprint similarity of N phonemes that meet the identification conditions in the broadband speech graph, and give an identification conclusion. The feature analysis method includes one or more of the formant trend feature analysis method, the formant frequency feature analysis method, and the intra-syllable transition sound feature analysis method.

[0037] Optionally, the system further comprises: a motion state detection module, which is used to obtain the motion state of the speaker according to the speech sample analysis; and the voiceprint identification module selects a feature analysis method according to the motion state of the speaker.

[0038] Optionally, the speaker's motion state includes but is not limited to one or more of still, uniform, accelerating, decelerating, straight, and curved.

[0039] Optionally, the data processing module may be voice processing software, or may be various voice workstations, voice analysis systems, and the like.

[0040] Optionally, the formant trend feature analysis method includes: analyzing the curve shape of the formant center line in the broadband spectrum, selecting at least one of the following characteristic phonemes to analyze the formant trend feature: monophthong finals, diphthong finals, and triphthong finals.

[0041] Optionally, the monophthongs include: “a”, “o”, “e”, “i”, “u”, “ü”.

[0042] Preferably, the monophthong finals are "u" and / or "e".

[0043] Optionally, the diphthong finals include at least one of the following: “ao”, “ai”, “ia”, “ue”.

[0044] Preferably, the diphthong finals are "ao" and / or "ai".

[0045] Optionally, the tri-vowel finals include at least one of the following: “iao”, “iou”, “uai”, “uei”.

[0046] Preferably, the tri-vowel final is "iao".

[0047] Optionally, the formant frequency characteristic analysis method includes: analyzing the frequency value, mean value, standard deviation and coefficient of variation of the characteristic phoneme formant, and judging whether the coefficient of variation CV1 is within a reasonable range.

[0048] Optionally, the characteristic phonemes include one or more of “a”, “o”, “e”, “i”, “u”, “ü”, “ao”, “ai”, “ia”, “ue”, “iao”, “iou”, “uai”, and “uei”.

[0049] Preferably, the characteristic phonemes are "a", "i", "e" and "ai".

[0050] Optionally, the range of the coefficient of variation CV1 is set to be less than 10%, which means that the formant frequency feature analysis item is considered to have passed, otherwise, the next phoneme identification is entered.

[0051] Optionally, the coefficient of variation calculation formula is as follows:

[0052]

[0053]

[0054]

[0055] Where σ is the standard deviation of the formant frequency, x1 and x2 are the formant frequencies of the selected characteristic phonemes in the speech specimen and speech sample, respectively. It is the average value of the formant frequency of the selected characteristic phonemes in the speech specimen and speech sample.

[0056] Preferably, x1 and x2 are average values ​​of the formant frequencies of the selected characteristic phonemes in the speech specimens and speech samples measured multiple times.

[0057] Optionally, the intra-syllable transition phonetic feature analysis method: determines whether the intra-syllable vowel-consonant transition segment morphological features of the characteristic phonemes of the speech specimen and the speech sample are consistent. If they are consistent, the authentication is passed; if not, proceed to the next phoneme identification.

[0058] Optionally, the characteristic phonemes include one or more of “an”, “en”, “in” and “un”.

[0059] Optionally, the characteristic phoneme is "an".

[0060] Optionally, the number of selected phonemes N≥10.

[0061] Preferably, the number of phonemes is 10≤N≤20.

[0062] The beneficial effects of this application include:

[0063] 1) The method and system for identity identification based on voiceprints in motion provided in this application, with respect to the motion state of the speaker, discovers through experimental comparison the features that change, remain basically unchanged, or only slightly change with the change of the motion state of the speaker, and uses the latter for identity identification, which greatly improves the efficiency and accuracy of identification.

[0064] 2) This application is applicable to voiceprint identification under various motion forms, without limiting the motion state of the speaker when recording the voiceprint, and even relaxes the restrictions on the recording conditions of the voice sample, so that the recording conditions of the voice sample do not need to rely on the voice sample, and only need to be recorded according to the standard procedure, which greatly improves the efficiency. The method and system of this application can select the appropriate feature analysis method for voice sample to perform voiceprint identification, and has a wide range of applications.

[0065] 3) This application provides a variety of feature extraction and analysis methods for motion states, including resonance peak trend feature analysis method, resonance peak frequency feature analysis method, and intra-syllable transition sound feature analysis method. One or more of them can be selected for analysis to ensure the accuracy of voiceprint identification.

[0066] 4) The method and system for identifying the identity of voiceprints in motion provided in this application require little data calculation and can be programmed online, so it is suitable for online use. The recognition mode can be automatically started by simply inputting the sample voice, sample voice and judgment conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of a method for identifying the identity of a voiceprint in motion in one embodiment of the present application;

[0068] Figure 2 The single-voice formant peak trend characteristics of the experimenter M1 at four movement speeds in one embodiment of the present application: (a) 0km / h, (b) 5km / h, (c) 10km / h, (d) 15km / h;

[0069] Figure 3 The diphthong formant trend characteristics of the experimenter M2 at four movement speeds in one embodiment of the present application: (a) 0 km / h, (b) 5 km / h, (c) 10 km / h, (d) 15 km / h;

[0070] Figure 4The triphthong formant trend characteristics of the experimenter W1 at four movement speeds in one embodiment of the present application: (a) xiao (iao), (b) jiao (iao);

[0071] Figure 5 In one embodiment of the present application, the experimenter W1 performed transition sounds within a syllable at four movement speeds: (a) yes (sh-i), (b) look (an). DETAILED DESCRIPTION

[0072] The present application is described in detail below with reference to embodiments, but the present application is not limited to these embodiments.

[0073] The present application provides a method for identifying an identity based on a voiceprint in motion, including:

[0074] S1: Obtain voice samples;

[0075] S2: converting the speech test material and speech sample into a broadband speech graph;

[0076] S3: Select a feature analysis method, select N phonemes that meet the identification conditions, perform feature identity analysis on the broadband speech graph, and obtain a voiceprint identification result.

[0077] The recording conditions of voice samples are uncertain. The reasons may include: the different movement states of the speaker, whether the speaker has any illness, whether the recording is open or secret, whether the recording device is obstructed (such as being hidden in a pocket), etc.

[0078] The existing technology usually directly uses commonly used phonemes to determine identity. Although it can also give an identification conclusion, the selection of phonemes here is usually random, and some phonemes will change in sound during movement, which reduces the accuracy of comparing the specimen with the sample, resulting in an erroneous identification conclusion.

[0079] In the present application, regardless of whether the motion state of the speaker in the sample is known, the method can be used for identification. Specifically, if the motion state of the speaker in the sample is not clear, the phonemes provided in the present application are used for identity identification, the accuracy of the identification result will be greatly improved, and under the condition of satisfying the number of phonemes (N), a conclusion can be directly drawn as to whether they are the same. If the motion state of the speaker in the sample is clear, the phonemes mentioned in the present application are directly selected, which can greatly improve the accuracy. Therefore, using the method of the present application, there is no need to consider the motion state of the speaker in the sample, and a unified standard procedure can be directly used for recording, thereby greatly improving efficiency.

[0080] In one embodiment, step S1 further includes: analyzing the movement state of the speaker, which can be obtained directly from the person submitting the material for examination, or can be determined based on auditory features in the material, such as the rhythm, fluency, coherence, environmental noise, and airflow of the speech sound.

[0081] The speaker's motion state includes but is not limited to stillness, uniform speed, acceleration, deceleration, straight line, and curve.

[0082] In one implementation, step S2 may use speech processing software such as Praat, Audition, Format Factory, etc. to convert speech materials and speech samples into broadband speech graphs, or may use various speech workstations, speech analysis systems, etc.

[0083] In one embodiment, the feature analysis method in step S3 may be selected from one or more of the following: formant trend feature analysis method, formant frequency feature analysis method, intra-syllable transition feature analysis method, such as Figure 1 shown.

[0084] The resonance peak trend characteristic analysis method may adopt one or more of the following:

[0085] ① Analysis of the formant trend of monophthong finals, where the monophthong finals are "a", "o", "e", "i", "u", and "ü".

[0086] In one embodiment, "a", "u", "i" and "e" are extracted from speech samples as research objects to observe whether there are differences in the direction characteristics of single vowel formant peaks of the same speaker at different movement speeds. Figure 2 The broadband speech graphs of the single vowel formant trends of the experimenter M1 at four movement speeds. As can be seen from the figure, the formant trend characteristics of the tested single vowels are relatively clear. For "u" and "e", the formant trend characteristics at different movement speeds are basically the same and are not affected by the speaker's movement speed. The formant trend characteristics of "a" and "i" have slight changes at different walking speeds, but do not show essential differences. Therefore, "u" and "e" are given priority for single vowel final formant trend analysis, especially when the speaker's movement state is unknown.

[0087] ② Analysis of the formant trend of diphthong finals, among which the diphthong finals are “ao”, “ai”, “ia” and “ue”.

[0088] In one embodiment, the diphthongs "ue", "ei", "ao", "ou", and "ai" of speech samples are extracted as research objects to observe whether there are differences in the diphthong formant trend characteristics of the same speaker at different movement speeds. Figure 3This is a broadband speechgram of the diphthong formant trends of experimenter M2 at four walking speeds. From the selected five formant trend characteristics, it can be seen that the formant trends of diphthongs uttered by the same person are different at different walking speeds. Specifically, the first formant trend of all diphthongs basically does not change with the change of movement speed, indicating that the first formant has the strongest anti-interference ability and the most stable characteristics; the trend characteristics of other formants are closely related to the diphthongs themselves. For example, the formant trends of the diphthongs "ao" and "ai" have slight differences at different movement speeds, but the overall change is not large, indicating that the pronunciation of these two diphthongs is very stable and can be used as an important reference for determining and denying identity; the high-order formant trends of "ue", "ei", and "ou" are greatly affected by the movement speed, and the characteristics vary a lot. If the test material and sample recording conditions are not clear, they should be used with caution.

[0089] ③Analysis on the trend of resonance peaks of triadic vowel finals, among which the triadic vowel finals are “iao”, “iou”, “uai” and “uei”.

[0090] In one embodiment, a speech sample triphthong "iao" is extracted as a research object to observe whether there are differences in the triphthong formant trend characteristics at different walking speeds for the same speaker. Figure 4 The broadband speech graphs of the triple vowel formant trends of the experimenter W1 at four movement speeds. As can be seen from the figure, the formant trends of the triple vowel "iao" at different movement speeds are basically the same. When the movement speed is the same, the formant trend characteristics of the triple vowel "iao" are very stable and are basically not affected by the initial consonant and the following syllable. Therefore, when the speaker's movement state is unknown, "iao" is preferred for the triple vowel final formant trend analysis.

[0091] In summary, the selection of the feature analysis method in step S3 can be determined according to the motion state of the speaker. For example, when the motion state of the speaker is known, the commonly used "a", "o", "e", "i", "u", "ü" are used as single vowel finals, "ao", "ai", "ia", "ue" are used as diphthong finals, and "iao", "iou", "uai", "uei" are used as triphthong finals; and when the motion state of the speaker is unknown, it is best to use "u" and "e" as single vowel finals, "ao" and "ai" as diphthong finals, and "iao" as triphthong finals, thereby greatly improving the accuracy of identification.

[0092] The formant frequency characteristic analysis method includes: analyzing the frequency value, average value, standard deviation and coefficient of variation of the characteristic phoneme formant, and judging whether the coefficient of variation CV1 is within a reasonable range.

[0093] In one embodiment, the coefficient of variation is calculated as follows:

[0094]

[0095]

[0096]

[0097] Where σ is the standard deviation of the formant frequency, x1 and x2 are the formant frequencies of the selected characteristic phonemes in the speech specimen and speech sample, respectively. It is the average value of the formant frequency of the selected characteristic phonemes in the speech specimen and speech sample.

[0098] In one embodiment, select Figure 2 , Figure 3 and Figure 4 The formant frequencies of vowels, diphthongs and triphthongs pronounced by speakers at different movement speeds were measured. The average value was taken after three measurements and statistical analysis was performed. The results are shown in Table 1 below.

[0099] Table 1 Formant frequencies of vowels of speakers at different walking speeds

[0100]

[0101]

[0102] Note: ① F1, F2, F3 and F4 represent the first, second, third and fourth resonance peaks respectively.

[0103] ②This resonance peak is not shown in the three-dimensional broadband spectrum.

[0104] Based on Table 1, it can be seen from the frequency value, mean value, standard deviation and coefficient of variation of vowel formants that the speed of movement has little effect on the frequency value of formants. Whether it is a single vowel, a diphthong or a triphthong, the coefficient of variation is less than 10% at different movement speeds, and even most of the formant frequency values ​​are less than 5%, which shows that the formant frequency has extremely high stability and is not easily affected by the speed of movement. It can be used as an important basis for identifying and denying identity. It should be noted that frequency measurement is highly subjective. Even with the help of the most advanced voice workstation or software, it is still impossible to eliminate the influence of the subjectivity of the identification personnel. It is only suitable as a verification method, and it is impossible to directly obtain the identification result of the identity by detecting the formant frequency of any phoneme. In fact, it is the change in the direction of the phoneme formant that causes the frequency of the sound to change accordingly. Therefore, it is necessary to use the formant frequency feature, and when using it, it is crucial to select the vowel. The straight formant in the vowel should be selected as much as possible for frequency measurement, such as "a", "i", "e", "ai", etc., to reduce the measurement error. Vowels containing "u" and "o" should be used with caution: on the one hand, since their first and second resonance peaks are generally concentrated in the low-frequency area, forming a spectral center of gravity, they are not easy to distinguish on a broadband spectrum and are prone to large measurement errors; on the other hand, the pronunciation of "u" and "o" is greatly affected by the shape of the oral cavity, and the direction of the resonance peaks is prone to change, which will further increase the frequency measurement error.

[0105] The intra-syllable transition phonetic feature analysis method determines whether the intra-syllable vowel-consonant transition segment morphological features of the characteristic phonemes of the speech test material and the speech sample are consistent. If they are consistent, the authentication is passed. If not, the next phoneme identification is entered.

[0106] In one embodiment, the transitional sound signs from "sh" to "i" in the speech sample "shi" and from "a" to "n" in the speech sample "kan" are extracted to study the morphological features of the consonant-vowel and vowel-consonant transition segments within the syllable. Figure 5 The following are broadband spectrograms of “shi” and “kan” uttered by experimenter W1 at different movement speeds. Figure 5 As can be seen in (a), there is no clear boundary between the "sh" and "i" pronounced by W1, which is a fuzzy transition, and the transition method does not change with the walking speed. However, as the walking speed increases, the transition sound direction of "sh" to "i" changes from a flat transition to a falling transition. Figure 5(b) It can be seen that at different movement speeds, the transition sign from "a" to "n" is in a descending direction, and the transition method is continuous transition. From the above examples, it can be seen that the morphological characteristics of the transition segment between consonants and vowels within a syllable will be affected by the walking speed, especially the morphological characteristics of the transition sign, which may be related to the airflow disturbance in the pronunciation channel during walking. The morphological characteristics of the transition segment between vowels and consonants within a syllable are basically not affected by the movement speed and can be used as an important reference for identifying and negating identity.

[0107] In one implementation, when N meets the identification conditions, an identification conclusion is given based on the voiceprint similarity of each feature.

[0108] In one implementation, 10≤N≤20. Too few phonemes may lead to inaccurate analysis results, while too many phonemes may affect analysis efficiency.

[0109] The present application also provides an identity identification system based on voiceprints in motion, including:

[0110] A data processing module, used for converting speech test materials and speech samples into broadband speech graphs;

[0111] The voiceprint identification module is used to select a feature analysis method, analyze the voiceprint similarity of N phonemes that meet the identification conditions in the broadband speech graph, and give an identification conclusion.

[0112] In one embodiment, the speaker's motion state includes but is not limited to one or more of still, uniform, accelerating, decelerating, straight, and curved.

[0113] In one embodiment, the system further includes: a motion state detection module for obtaining the motion state of the speaker according to the speech sample analysis; and the voiceprint identification module selects a feature analysis method according to the motion state of the speaker.

[0114] In one implementation, the data processing module may be voice processing software, or may be various voice workstations, voice analysis systems, and the like.

[0115] In one embodiment, the voiceprint identification module selects a feature analysis method including one or more of a formant trend feature analysis method, a formant frequency feature analysis method, and an intra-syllable transition sound feature analysis method.

[0116] In one embodiment, the formant trend feature analysis method includes: analyzing the curve shape of the formant center line in the broadband spectrum, selecting at least one of the following characteristic phonemes to analyze the formant trend feature: monophthong finals, diphthong finals, and triphthong finals.

[0117] In one implementation, the monophthongs include: “a”, “o”, “e”, “i”, “u”, and “ü”.

[0118] As a preferred embodiment, the monophthong finals are "u" and / or "e".

[0119] In one implementation, the diphthong finals include at least one of the following: “ao”, “ai”, “ia”, “ue”.

[0120] As a preferred embodiment, the diphthong finals are "ao" and / or "ai".

[0121] In one implementation, the tri-vowel finals include at least one of the following: “iao”, “iou”, “uai”, “uei”.

[0122] As a preferred embodiment, the tri-vowel final is "iao".

[0123] In one embodiment, the formant frequency characteristic analysis method includes: analyzing the frequency value, mean value, standard deviation and coefficient of variation of the characteristic phoneme formant, and determining whether the coefficient of variation CV1 is within a reasonable range.

[0124] In one implementation, the characteristic phonemes include one or more of “a”, “o”, “e”, “i”, “u”, “ü”, “ao”, “ai”, “ia”, “ue”, “iao”, “iou”, “uai”, and “uei”.

[0125] As a preferred embodiment, the characteristic phonemes are "a", "i", "e" and "ai".

[0126] In one implementation, the range of the coefficient of variation CV1 is set to be less than 10%, which means that the formant frequency feature analysis item is considered to have passed, otherwise, the next phoneme identification is entered.

[0127] As a preferred embodiment, the coefficient of variation calculation formula is as follows:

[0128]

[0129]

[0130]

[0131] Where σ is the standard deviation of the formant frequency, x1 and x2 are the formant frequencies of the selected characteristic phonemes in the speech specimen and speech sample, respectively. It is the average value of the formant frequency of the selected characteristic phonemes in the speech specimen and speech sample.

[0132] In one implementation, x1 and x2 are average values ​​of the formant frequencies of the selected characteristic phonemes in the speech specimens and the speech samples measured multiple times.

[0133] In one embodiment, the intra-syllable transition phonetic feature analysis method determines whether the intra-syllable vowel-consonant transition segment morphological features of the characteristic phonemes of the speech sample and the speech sample are consistent. If they are consistent, the authentication is passed; if not, the next phoneme identification is entered.

[0134] In one implementation, the characteristic phonemes include one or more of “an”, “en”, “in”, and “un”.

[0135] As a preferred embodiment, the characteristic phoneme is "an".

[0136] In one implementation, the number of selected phonemes N≥10.

[0137] As a preferred embodiment, the number of phonemes is 10≤N≤20.

[0138] The above are only a few embodiments of the present application and do not constitute any form of limitation to the present application. Although the present application is disclosed as above with preferred embodiments, it is not intended to limit the present application. Any technician familiar with the profession, without departing from the scope of the technical solution of the present application, using the technical content disclosed above to make slight changes or modifications are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A method for identifying the identity of a voiceprint in motion, characterized in that: The method includes: Acquiring speech samples, and converting the speech samples and speech samples into broadband speech graphs; Select a feature analysis method, select N phonemes, and when N meets the identification conditions, perform feature consistency analysis on the broadband speech graph to obtain a voiceprint identification result; wherein the feature analysis method adopts one or more of the following: resonance peak trend feature analysis method, resonance peak frequency feature analysis method, and intra-syllable transition sound feature analysis method.

2. The method for identifying identity based on voiceprint in motion according to claim 1, characterized in that: The method further comprises: after acquiring the speech sample, analyzing the motion state of the speaker; the feature analysis method is selected according to the motion state of the speaker; Preferably, the speaker's motion state includes at least one of the following: still, uniform, accelerating, decelerating, straight, and curved.

3. The method for identifying identity based on voiceprint in motion according to claim 1, characterized in that: The voice test material and voice sample are converted into a broadband speech graph using one or more of a voice processing software, a voice workstation, and a voice analysis system.

4. The method for identifying identity based on voiceprint in motion according to claim 1, characterized in that: The formant trend feature analysis method comprises: analyzing the curve shape of the formant center line in the broadband spectrum, selecting at least one of the following characteristic phonemes to analyze the trend feature of the formant: monophthong finals, diphthong finals, triphthong finals; Preferably, the monophthongs include at least one of the following: "a", "o", "e", "i", "u", "ü"; Preferably, the monophthong final is selected from at least one of the following: "u", "e"; Preferably, the diphthong finals include at least one of the following: "ao", "ai", "ia", "ue"; Preferably, the diphthong finals are selected from at least one of the following: "ao", "ai"; Preferably, the tri-vowel finals include at least one of the following: "iao", "iou", "uai", "uei"; Preferably, the tri-vowel final is "iao".

5. The method for identifying identity based on voiceprint in motion according to claim 1, characterized in that: The formant frequency characteristic analysis method includes: analyzing the frequency value, average value, standard deviation and coefficient of variation of the characteristic phoneme formant to determine whether the coefficient of variation CV1 is within a reasonable range; Preferably, the characteristic phonemes include at least one of: "a", "o", "e", "i", "u", "ü", "ao", "ai", "ia", "ue", "iao", "iou", "uai", "uei"; Preferably, the characteristic phoneme is selected from at least one of the following: "a", "i", "e", "ai"; Preferably, the range of the coefficient of variation CV1 is set to be less than 10%, which means that the formant frequency feature analysis item is considered to have passed, otherwise, the next phoneme identification is entered; Preferably, the coefficient of variation calculation formula is as follows: Where σ is the standard deviation of the formant frequency, x1 and x2 are the formant frequencies of the selected characteristic phonemes in the speech specimen and speech sample, respectively. is the average value of the formant frequency of the selected characteristic phonemes in the speech test material and speech sample; Preferably, x1 and x2 are average values ​​of the formant frequencies of the selected characteristic phonemes in the speech specimens and speech samples measured multiple times.

6. The method for identifying identity based on voiceprint in motion according to claim 1, characterized in that: The intra-syllable transition sound feature analysis method: determines whether the intra-syllable vowel-consonant transition segment morphological features of the characteristic phonemes of the speech test material and the speech sample are consistent. If they are consistent, the authentication is passed. If not, the next phoneme identification is entered; Preferably, the characteristic phonemes include at least one of the following: "an", "en", "in", "un"; Preferably, the characteristic phoneme is "an".

7. The method for identifying identity based on voiceprint in motion according to claim 1, characterized in that: The number of selected phonemes N ≥ 10; Preferably, the number of phonemes is 10≤N≤20.

8. A system for identifying identity based on voiceprints in motion, characterized in that: The system includes: A data processing module, used for converting speech test materials and speech samples into broadband speech graphs; The voiceprint identification module is used to select a feature analysis method, analyze the voiceprint similarity of N phonemes that meet the identification conditions in the broadband speech graph, and give an identification conclusion; the feature analysis method includes at least one of the following: formant trend feature analysis method, formant frequency feature analysis method, and intra-syllable transition sound feature analysis method.

9. The identity verification system based on voiceprint in motion according to claim 8, characterized in that: The system also includes: The motion state detection module is used to obtain the motion state of the speaker based on the speech sample analysis; The voiceprint identification module selects a feature analysis method according to the movement state of the speaker; Preferably, the speaker's motion state includes at least one of the following: still, uniform, accelerating, decelerating, straight, and curved.

10. The identity verification system based on voiceprint in motion according to claim 8, characterized in that: The formant trend feature analysis method comprises: analyzing the curve shape of the formant center line in the broadband spectrum, selecting at least one of the following characteristic phonemes to analyze the trend feature of the formant: monophthong finals, diphthong finals, triphthong finals; Preferably, the monophthongs include at least one of the following: "a", "o", "e", "i", "u", "ü"; Preferably, the monophthong final is selected from at least one of the following: "u", "e"; Preferably, the diphthong finals include at least one of the following: "ao", "ai", "ia", "ue"; Preferably, the diphthong finals are selected from at least one of the following: "ao", "ai"; Preferably, the tri-vowel finals include at least one of the following: "iao", "iou", "uai", "uei"; Preferably, the tri-vowel final is "iao"; Preferably, the formant frequency characteristic analysis method includes: analyzing the frequency value, mean value, standard deviation and coefficient of variation of the characteristic phoneme formant to determine whether the coefficient of variation CV1 is within a reasonable range; Preferably, the characteristic phonemes include at least one of the following: "a", "o", "e", "i", "u", "ü", "ao", "ai", "ia", "ue", "iao", "iou", "uai", "uei"; Preferably, the characteristic phoneme is selected from at least one of the following: "a", "i", "e", "ai"; Preferably, the range of the coefficient of variation CV1 is set to be less than 10%, which means that the formant frequency feature analysis item is considered to have passed, otherwise, the next phoneme identification is entered; Preferably, the coefficient of variation calculation formula is as follows: Where σ is the standard deviation of the formant frequency, x1 and x2 are the formant frequencies of the selected characteristic phonemes in the speech sample and speech material, respectively. is the average value of the formant frequency of the selected characteristic phonemes in the speech test material and speech sample; Preferably, x1 and x2 are average values ​​of the formant frequencies of the selected characteristic phonemes in the speech test materials and speech samples measured multiple times; Preferably, the intra-syllable transition sound feature analysis method: determines whether the intra-syllable vowel-consonant transition segment morphological features of the characteristic phonemes of the speech test material and the speech sample are consistent, if consistent, the authentication is passed, if not, proceed to the next phoneme identification; Preferably, the characteristic phonemes include at least one of the following: "an", "en", "in", "un"; Preferably, the characteristic phoneme is "an"; Preferably, the number of phonemes N ≥ 10; Preferably, the number of the phonemes is 10≤N≤20.