Method for early recognition of parkinsonian dysarthria based on voiceprint features

Abstract

The application discloses a Parkinson disease dysarthria early identification method based on voiceprint features and belongs to the technical field of medical diagnosis. The original speech data of a to-be-identified object is subjected to speech analysis and speech recognition, and an initial feature set related to voiceprints in a pronunciation deviation and a segment is extracted, target features with significant discriminability for Parkinson disease dysarthria identification are screened out, a dysarthria identification result of the to-be-identified object is obtained through a model, accurate vowel and consonant segment extraction and pronunciation deviation identification are realized, phoneme segment accurate cutting is realized by combining the association of a text sequence and speech data, the quantization determination of the pronunciation deviation is realized, the key dimension of the dysarthria identification is reserved through feature extraction, the early signals of Parkinson disease dysarthria are captured, the high-risk groups are marked, the early warning result is output in combination with the joint risk value calculation, the problem that the prodromal symptoms are not obvious is solved, the review interval is dynamically adjusted, and the whole-process management from identification to monitoring is realized.

Description

Technical Field

[0001] This invention relates to the field of medical diagnostic technology, and in particular to an early identification method for dysarthria in Parkinson's disease based on voiceprint features. Background Technology

[0002] Currently, the clinical diagnosis of dysarthria in Parkinson's disease mainly relies on the subjective assessment of physicians, such as using the speech function scoring items in the Standardized Parkinson's Disease Rating Scale (SDR) to make a qualitative judgment by observing the patient's pronunciation, speech rate, and fluency. However, this subjective assessment method is easily affected by factors such as the physician's clinical experience and the assessment environment, and has problems such as strong subjectivity in identification, low sensitivity in early signal capture, and poor consistency of assessment results, making it difficult to detect subtle articulation abnormalities in the prodromal or early stages.

[0003] Meanwhile, some methods focus only on single-dimensional speech features, making it difficult to fully capture the complex pathological manifestations of dysarthria; some methods use speech collection content that lacks specificity, resulting in insufficient feature extraction and limited recognition accuracy; at the same time, most existing methods focus on identifying patients who have already shown obvious symptoms of dysarthria, making it difficult to achieve accurate screening and early warning of high-risk groups in the prodromal stage, and failing to meet the needs of early clinical intervention. Summary of the Invention

[0004] The purpose of this invention is to provide an early identification method for dysarthria in Parkinson's disease based on voiceprint features, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an early identification method for dysarthria in Parkinson's disease based on voiceprint features, comprising the following steps:

[0006] Acquire the raw speech data of the object to be identified and preprocess it;

[0007] Speech analysis and speech recognition are performed on the preprocessed speech data to determine the effective speech segments and semantic information corresponding to the speech content. At the same time, vowel segments and consonant segments are extracted from the speech data, and pronunciation deviations of consonants and vowels are identified.

[0008] Speech processing is performed on valid speech segments to extract an initial feature set related to the voiceprint within the segment. The initial feature set includes the time-domain features, frequency-domain features, nonlinear features, vowel-related parameters, and language fluency index of the voiceprint.

[0009] Feature selection is performed on the initial feature set to screen out target features that have significant discriminative power in identifying articulation disorders in Parkinson's disease. The target features include vowel space area, vowel articulation index, speech pause duration, speech rate variation, voice roughness, and speech speed.

[0010] The target features are input into a pre-trained Parkinson's disease dysarthria identification model. The model calculates the dysarthria identification results of the subject to be identified, thus completing the identification of high-risk groups in the prodromal stage and achieving early identification.

[0011] Furthermore, the process of extracting vowel and consonant segments from the speech data and identifying pronunciation deviations between consonants and vowels includes:

[0012] Based on the text sequence obtained from speech recognition, the pronunciation periods of corresponding vowels and consonants in the speech data are determined, and vowel segments and consonant segments are extracted.

[0013] Calculate the formant frequencies of vowel segments, establish the acoustic spatial distribution of vowels, and determine the standard position range of each vowel in acoustic space;

[0014] Consonant acoustic features are extracted from consonant segments, including plosive duration, friction intensity, and voicing degree, and the standard feature range for each consonant is determined.

[0015] In determining the standard position interval of vowels and the standard feature range of consonants, based on the speech data of healthy people, a 95% confidence interval is obtained through statistical analysis, which serves as the standard position interval and standard feature range. The standard range is updated periodically based on new speech data samples from healthy people, and the number of speech data samples from healthy people is not less than 1,000.

[0016] The deviation values ​​of the actual acoustic spatial position of the vowel segment to be identified from the standard position range are compared with the deviation values ​​of the actual acoustic features of the consonant segment from the standard feature range. When the deviation value is greater than the preset deviation threshold, it is determined that there is a pronunciation deviation, and the deviation type and degree are recorded.

[0017] Furthermore, the standard range is updated periodically based on new voice data samples from healthy individuals, including:

[0018] The voice data samples of the new healthy population are divided according to age to obtain multiple age sample groups; each age sample group is a sample group composed of voice data samples of healthy people belonging to the same age range.

[0019] Vowel features and auxiliary features were extracted from the speech data samples of each age group to obtain the analytical vowel features and analytical auxiliary features;

[0020] Based on the analysis of vowel features, the first reference position interval of vowels in the age sample group was determined.

[0021] Based on the analysis of auxiliary features, the range of the first reference feature for consonants in the corresponding age sample group is determined.

[0022] The degree of difference between the first reference position interval and the historical standard position interval of the same vowel in the same age interval is obtained by comparing the differences between the position intervals.

[0023] The degree of difference in the feature range is obtained by comparing the first reference feature range with the historical standard feature range of the same consonant in the same age range.

[0024] When the degree of difference in position interval or the degree of difference in feature range exceeds the corresponding preset difference constraint, the first reference position interval or the first reference feature range is regarded as the expected position interval of the corresponding vowel or the expected feature range of the corresponding consonant in the current age sample group.

[0025] When the degree of difference in position intervals or the degree of difference in feature range does not exceed the corresponding preset difference constraint, the expected position interval of the corresponding vowel in the current age sample group is obtained by weighted averaging the first reference position interval and the historical standard position interval, or the expected feature range of the corresponding consonant in the current age sample group is obtained by weighted averaging the first reference feature range and the historical standard feature range.

[0026] The sample statistics of each age group were carried out according to different regions to determine the sample proportion of multiple regions.

[0027] Obtain the common vowel position range and common consonant feature range of a region;

[0028] For all general vowel position intervals that differ from the historical standard position intervals of the same vowel in the same region beyond the preset difference constraint, the expected position intervals of the corresponding vowels in the age sample group are adjusted based on the regional sample proportion to obtain new standard position intervals.

[0029] The expected feature range of the corresponding consonant in the age sample group is adjusted by taking into account the regional sample proportion and the degree of difference between the historical standard feature range of all consonants in the same region and the general consonant feature range that exceeds the preset difference constraint, so as to obtain a new standard feature range.

[0030] Furthermore, the process of extracting vowel-related parameters and language fluency indicators includes:

[0031] For each vowel segment, a two-dimensional acoustic space is constructed based on the formant frequencies, the distance between different vowels is calculated, and the area of ​​the vowel space is determined.

[0032] In calculating the vowel space area, if the object to be identified has some missing vowel pronunciations, the interpolation method is used to supplement the standard position of the missing vowel based on the existing vowel acoustic features, and then the vowel space area is calculated. The interpolation weight is determined according to the acoustic correlation between the missing vowel and the existing vowel.

[0033] The clarity, stability, and duration of vowel pronunciation are calculated, and a vowel pronunciation index is obtained by weighted summation.

[0034] The effective speech segments are divided into time segments, pauses in the speech signal are identified, the duration of each pause is counted, and the speech pause duration distribution is obtained.

[0035] The number of written syllables corresponding to speech per unit time is calculated as the speech rate value, and the rate of change of speech rate in different time periods is calculated using the sliding window method as the speech rate change index.

[0036] Furthermore, the clarity of vowel pronunciation is calculated, including:

[0037] Obtain the preset cepstral features of vowels; the preset cepstral features include cepstral peak height, number of cepstral peaks, and cepstral peak position;

[0038] The deviation analysis is performed on the cepstral peak position of the vowel and the formant position of the corresponding standard vowel and the preset theoretical position of the corresponding vowel category, respectively, to obtain the first position deviation status and the second position deviation status.

[0039] When both the first position deviation condition and the second position deviation condition are less than the corresponding position deviation threshold, the larger value between the first position deviation condition and the second position deviation condition is selected as the reference position deviation.

[0040] Otherwise, the average of the first and second position deviation conditions shall be regarded as the reference position deviation;

[0041] The clarity of vowel pronunciation is calculated using cepstral peak height, cepstral peak number, and reference position deviation.

[0042] Further, the process of extracting sound roughness and pronunciation speed features includes:

[0043] Perform a short-time Fourier transform on the effective speech segment to obtain the spectrum, and calculate the energy ratio of harmonic components to noise components in the spectrum;

[0044] The degree of irregular fluctuation in the speech signal is analyzed, and the quantification value of the sound roughness is obtained by weighting the energy ratio and the degree of fluctuation.

[0045] The average pronunciation speed is calculated by counting the total number of syllables in the text sequence corresponding to the speech segment and combining it with the total duration of the speech segment.

[0046] The transition duration between adjacent syllables during continuous pronunciation is analyzed, and the standard deviation of the transition duration is calculated as a characteristic of pronunciation speed stability. The average pronunciation speed and the speed stability characteristic are used together as characteristics related to pronunciation speed.

[0047] Furthermore, the screening process for target features, including vowel space area, vowel articulation index, duration of speech pauses, speech rate variation, harshness of voice, and pronunciation speed, includes:

[0048] The t-test method was used to calculate the significant differences in vowel space area, vowel articulation index, speech pause duration, speech rate variation, voice roughness, and speech speed between Parkinson's disease dysarthria samples and healthy population samples. Features with p-values ​​less than 0.05 were retained to obtain a preliminary feature set.

[0049] A feature importance evaluation model was constructed, with the accuracy of speech disorder recognition as the objective function. The importance score of each feature in the preliminary feature set was calculated using the random forest algorithm.

[0050] Sort the features by importance score from high to low, select the top N features, and combine them with time domain, frequency domain and nonlinear features to form the target features. N is preset according to the model complexity, and the preset range of N is 10-20.

[0051] When selecting the top N features by score, vowel space area, vowel articulation index, sound roughness, and articulation speed are mandatory features.

[0052] Furthermore, the process of identifying high-risk individuals in the prodromal stage by combining characteristics such as voice roughness and speech speed includes:

[0053] Based on samples of patients with dysarthria in the preclinical stage of Parkinson's disease, high-risk threshold ranges for voice roughness and speech speed were determined.

[0054] The quantification value of the roughness of the voice and the characteristic value of the pronunciation speed of the subject to be identified are judged. If they both fall into the corresponding high-risk threshold range, and the identification results of other target features are suspected articulation disorders, then they are marked as high-risk groups in the prodromal stage.

[0055] For high-risk groups, a joint risk value is calculated based on the roughness of the voice and the speed of speech. The joint risk value is obtained by weighted summation of the standardized values ​​of the two. When the joint risk value is greater than a preset risk threshold, a high-risk warning result is output.

[0056] Furthermore, after outputting the high-risk warning results, it also includes:

[0057] Extract the temporal variation trends of voice roughness and pronunciation speed characteristics of high-risk groups, combine them with historical voice data to analyze the deterioration rate of the characteristics, and shorten the re-examination interval when the deterioration rate exceeds a preset rate threshold.

[0058] Furthermore, the acquired raw speech data must include preset speech disorder sensitive sentences, which cover high-frequency vowels, consonant combinations, and content with different speech rates to ensure the effective extraction of vowel segments and consonant segments and the accurate calculation of related parameters.

[0059] Compared with the prior art, the beneficial effects of the present invention are:

[0060] 1. This invention collects raw speech data by pre-processing it using preset sensitive phrases for articulation disorders. Combined with precise extraction of vowel and consonant segments and identification of pronunciation deviations, it improves the targeting of early identification and the quality of basic data. The preset sensitive phrases cover high-frequency vowels, consonant combinations, and different speech rate requirements, which can accurately capture pronunciation scenarios where articulation disorders are prone to abnormalities. By associating text sequences with speech data, it achieves precise extraction of phoneme segments. Combined with multi-dimensional acoustic features such as formant frequency and plosive duration, it compares with the standard range established by a large sample of healthy people to achieve quantitative judgment of pronunciation deviations.

[0061] 2. This invention constructs a target feature set that combines comprehensiveness and discriminativeness by comprehensively extracting multi-dimensional speech features and employing a scientific feature selection strategy. The feature extraction process covers vowel-related parameters, speech fluency indicators, and features related to sound quality and pronunciation speed, thus characterizing the core manifestations of articulation disorders from multiple dimensions. By eliminating features without discriminativeness and quantifying feature importance using the random forest algorithm, combined with the setting of mandatory features and dimensional control, the final target feature set avoids redundant information interference while retaining the key dimensions for articulation disorder identification. This allows for a more comprehensive and accurate capture of early signals of Parkinson's disease articulation disorders, improving the sensitivity and specificity of the model's recognition and effectively solving the problem of low recognition accuracy caused by the one-sidedness and insufficient discriminativeness of traditional feature systems.

[0062] 3. This invention achieves early identification of dysarthria and precise management of high-risk groups by constructing a precise identification model and a dynamic monitoring mechanism for high-risk groups in the prodromal phase. Optimized target features are input into a pre-trained identification model, which can quickly output dysarthria identification results. For high-risk groups in the prodromal phase, high-risk threshold ranges for core features are set based on clinical samples. High-risk groups are marked through dual screening, and warning results are output by combining joint risk values. This solves the technical problem of unclear prodromal symptoms and difficulty in identification using traditional methods. By analyzing the temporal change trend and deterioration rate of high-risk group characteristics, the re-examination interval is dynamically adjusted, achieving full-process management from identification to monitoring. Attached Figure Description

[0063] Fig. 1 This is a schematic diagram of the overall method flow of the present invention;

[0064] Fig. 2 This is a schematic diagram of the feature extraction and screening process of the present invention;

[0065] Fig. 3 This is a schematic diagram of the high-risk population identification process of the present invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Please see Figs. 1-3 The present invention provides the following technical solutions:

[0068] An early identification method for dysarthria in Parkinson's disease based on voiceprint features includes the following steps:

[0069] The original speech data of the object to be identified is acquired and preprocessed. The acquired original speech data must contain pre-defined speech disorder sensitive sentences. The sentences cover high-frequency vowels, consonant combinations and content with different speech speed requirements, such as "eight hundred soldiers rush to the north slope" and "red phoenix and pink phoenix", to ensure the effective extraction of vowel segments and consonant segments and the accurate calculation of related parameters.

[0070] Speech analysis and speech recognition are performed on the preprocessed speech data to determine the effective speech segments and semantic information corresponding to the speech content. At the same time, vowel segments and consonant segments are extracted from the speech data, and pronunciation deviations of consonants and vowels are identified.

[0071] Speech processing is performed on valid speech segments to extract an initial feature set related to the voiceprint within the segment. The initial feature set includes the time-domain features, frequency-domain features, nonlinear features, vowel-related parameters, and speech fluency index of the voiceprint.

[0072] Feature selection was performed on the initial feature set to identify target features that have significant discriminative power in identifying dysarthria in Parkinson's disease. The target features include vowel space area, vowel articulation index, speech pause duration, speech rate variation, voice roughness, and speech speed.

[0073] The target features are input into a pre-trained Parkinson's disease dysarthria identification model. The model calculates the dysarthria identification results of the subject to be identified, thus completing the identification of high-risk groups in the prodromal stage and achieving early identification.

[0074] In the above embodiments, by pre-setting sensitive sentences for articulation disorders that cover high-frequency vowels, consonant combinations, and different speech rate requirements, such sensitive sentences can accurately cover pronunciation scenarios where articulation disorders are prone to abnormalities, ensuring the complete extraction of vowel and consonant segments, avoiding feature loss due to monotonous speech content or simple pronunciation, significantly reducing the interference of irrelevant information on the recognition results, ensuring the accuracy and reliability of the entire recognition process, and solving the technical problems of insufficient feature extraction and low recognition accuracy caused by the lack of targeted content in traditional speech acquisition.

[0075] The process of extracting vowel and consonant segments from speech data and identifying pronunciation deviations between consonants and vowels includes:

[0076] Based on the text sequence obtained from speech recognition, the pronunciation time periods of corresponding vowels (a, o, e, i, u) and consonants (b, p, m, f, etc.) in the speech data are determined, and vowel segments and consonant segments are extracted.

[0077] Calculate the formant frequencies of vowel segments, establish the acoustic spatial distribution of vowels, and determine the standard position range of each vowel in acoustic space;

[0078] Consonant acoustic features are extracted from consonant segments. These features include plosive duration, friction intensity, and voicing degree. The standard feature range for each consonant is then determined.

[0079] In determining the standard position intervals of vowels and the standard feature ranges of consonants, based on the speech data of healthy individuals, a 95% confidence interval is obtained through statistical analysis, which serves as the standard position interval and standard feature range. The standard ranges are updated periodically based on new speech data samples from healthy individuals, with a sample size of no less than 1,000 cases.

[0080] The deviation values ​​of the actual acoustic spatial position of the vowel segment to be identified from the standard position range are compared with the deviation values ​​of the actual acoustic features of the consonant segment from the standard feature range. When the deviation value is greater than the preset deviation threshold, it is determined that there is a pronunciation deviation, and the deviation type and degree are recorded.

[0081] In the above embodiments, the precise extraction of vowel and consonant segments is achieved through the correspondence between text sequences and speech data. Formant frequencies are calculated for vowels and acoustic spatial distribution is established. Key acoustic features such as plosive duration are extracted for consonants, thus characterizing the acoustic essence of pronunciation from different dimensions. Multi-dimensional feature analysis can more comprehensively capture pronunciation abnormalities caused by articulation disorders. At the same time, the standard range is determined based on a large sample of healthy population data to ensure the objectivity and universality of the standard. The design of regularly updating the standard range adapts to the dynamic changes in the acoustic characteristics of the population and avoids recognition bias caused by the solidification of the standard. By comparing the actual features with the standard range, the deviation is quantified, and the pronunciation deviation is accurately determined and the type is recorded, providing a core basis for subsequent feature extraction and model recognition, and improving the sensitivity of early signal capture of articulation disorders.

[0082] The standard range is updated periodically based on new voice data samples from healthy individuals, including:

[0083] The voice data samples of the new healthy population are divided according to age to obtain multiple age sample groups; each age sample group is a sample group composed of voice data samples of healthy people belonging to the same age range.

[0084] Vowel features and auxiliary features were extracted from the speech data samples of each age group to obtain the analytical vowel features and analytical auxiliary features;

[0085] Based on the analysis of vowel features, the first reference position interval of vowels in the age sample group was determined.

[0086] Based on the analysis of auxiliary features, the range of the first reference feature for consonants in the corresponding age sample group is determined.

[0087] The degree of difference between the first reference position interval and the historical standard position interval of the same vowel in the same age interval is obtained by comparing the differences between the position intervals.

[0088] The degree of difference in the feature range is obtained by comparing the first reference feature range with the historical standard feature range of the same consonant in the same age range.

[0089] When the degree of difference in position interval or the degree of difference in feature range exceeds the corresponding preset difference constraint, the first reference position interval or the first reference feature range is regarded as the expected position interval of the corresponding vowel or the expected feature range of the corresponding consonant in the current age sample group.

[0090] When the degree of difference in position intervals or the degree of difference in feature range does not exceed the corresponding preset difference constraint, the expected position interval of the corresponding vowel in the current age sample group is obtained by weighted averaging the first reference position interval and the historical standard position interval, or the expected feature range of the corresponding consonant in the current age sample group is obtained by weighted averaging the first reference feature range and the historical standard feature range.

[0091] The sample statistics of each age group were carried out according to different regions to determine the sample proportion of multiple regions.

[0092] Obtain the common vowel position range and common consonant feature range of a region;

[0093] For all general vowel position intervals that differ from the historical standard position intervals of the same vowel in the same region beyond the preset difference constraint, the expected position intervals of the corresponding vowels in the age sample group are adjusted based on the regional sample proportion to obtain new standard position intervals.

[0094] The expected feature range of the corresponding consonant in the age sample group is adjusted by taking into account the regional sample proportion and the degree of difference between the historical standard feature range of all consonants in the same region and the general consonant feature range that exceeds the preset difference constraint, so as to obtain a new standard feature range.

[0095] The standard range includes the standard position interval of vowels in acoustic space and the standard characteristic range of consonants (such as the reasonable range of values ​​for plosive duration, friction intensity, and degree of voicing).

[0096] An age sample cluster is a collection of voice data samples from healthy individuals belonging to the same age range. For example, voice data samples from new healthy individuals can be divided into multiple age ranges such as 20-29 years old, 30-39 years old, 40-49 years old, and 50-59 years old. Voice data samples from healthy individuals within each age range constitute an age sample cluster.

[0097] The analysis of vowel features involved identifying multiple formant frequencies from speech data samples of all healthy individuals included in the age sample group.

[0098] The consonant features were analyzed as acoustic features of consonants determined from speech data samples of all healthy individuals included in the age sample group.

[0099] The first reference position interval is determined based on the analysis of vowel features. The reference interval of vowel position in acoustic space in the current age sample group reflects the distribution of acoustic features of vowel pronunciation in new healthy people belonging to different age intervals.

[0100] Understandably, the first reference position interval can be determined by first calculating the mean and standard deviation of all formant frequencies in the corresponding age sample group, and then using the mean of the formant frequencies as the center, combined with the corresponding standard deviation.

[0101] The first reference feature range is determined based on the analysis of auxiliary features, and is the reference range of consonant pronunciation features in the current age sample group.

[0102] Understandably, the first reference feature range can be determined by first calculating the mean and standard deviation of the consonant acoustic features in the corresponding age sample group, and then using the mean of the consonant acoustic features as the center, combined with the corresponding standard deviation.

[0103] The historical standard position interval is the standard position interval of vowels corresponding to each age interval obtained from the previous historical update.

[0104] The degree of positional interval difference is a measure of the difference between the first reference positional interval and the historical standard positional interval of the same vowel in the same age interval, which can be obtained by calculating the difference between the mean values ​​of the two intervals.

[0105] The historical standard feature range is the standard feature range of the consonant acoustic features corresponding to each age range obtained from the previous historical update.

[0106] The degree of difference in feature range is a measure of the magnitude of the difference between the first reference feature range and the historical standard feature range of the same consonant in the same age range, which can be obtained by calculating the difference between the means of the two ranges.

[0107] The preset difference constraint is a threshold set in advance based on the analysis of changes in a large amount of speech data to determine whether the degree of difference in positional intervals or feature ranges is significant. When the degree of difference exceeds the preset difference constraint, it is considered that the pronunciation features have changed significantly, and a new reference interval or feature range needs to be used as the expected value.

[0108] The expected position interval is determined based on the comparison results between the degree of difference between position intervals and the preset difference constraints. It is the reference interval for the position of the corresponding vowel in acoustic space in the current age sample group.

[0109] If the difference between the location intervals does not exceed the corresponding preset difference constraints, the expected location interval can be obtained by directly weighting the first reference location interval and the historical standard location interval. The weights assigned to the two intervals can be obtained by solving the matrix constructed by pairwise comparison and scoring using the analytic hierarchy process, with values ​​ranging from (0,1).

[0110] For example, assuming the vowel / a / in the age sample group of 20-29 years old has a position interval difference of 20Hz, which does not exceed the preset difference constraint of 50Hz, and the weight assigned to the first reference position interval is 0.6, and the weight assigned to the historical standard position interval is 0.4.

[0111] At this point, a weighted average is taken between the first reference position interval [300Hz, 400Hz] and the historical standard position interval [340Hz, 400Hz] to obtain the expected position interval. That is, the expected position range is .

[0112] The expected feature range is determined based on the comparison results between the degree of difference in feature range and the preset difference constraint, and is the reference range of pronunciation features of corresponding consonants in the current age sample group.

[0113] If the difference in the feature range does not exceed the corresponding preset difference constraint, the expected feature range can be calculated by directly weighting the first reference feature range and the historical standard feature range. The weights assigned to the two intervals can be obtained by solving the matrix constructed by pairwise comparison and scoring using the analytic hierarchy process, and the values ​​are all in the range of (0,1).

[0114] A region is a different area divided according to geographical regions. For example, a region can include East China, Central China, North China, South China, Southwest China, Northwest China, and Northeast China.

[0115] The regional sample proportion refers to the proportion of voice data samples from different regions in each age group to the total number of samples in that age group.

[0116] The universal vowel position range is the general position range of vowels in acoustic space obtained from statistical analysis of speech data of healthy people in a specific region. This range reflects the general characteristics of vowel pronunciation in that region and takes into account the influence of regional culture and language habits on the pronunciation of different vowels.

[0117] The general range of consonant features refers to the common range of values ​​for consonant pronunciation features (such as plosive duration, friction intensity, and voicing degree) derived from statistical analysis of speech data of healthy people in a specific region, reflecting the influence of regional factors on consonant pronunciation.

[0118] The new standard position interval is a new reference interval used to replace the original historical standard position interval. It is obtained by adjusting the expected position interval of the corresponding vowels in the age sample group by using the general vowel position interval and the regional sample proportion.

[0119] The new standard position interval can be calculated by directly weighting the general vowel position intervals that differ from the historical standard position intervals of the same vowel in the same region by more than the preset difference constraint, and the expected position intervals of the corresponding vowel in the corresponding age sample group.

[0120] The weights assigned to the general vowel position intervals where the difference from the historical standard position intervals of the same vowel in the same region exceeds the preset difference constraints, and the expected position intervals of the corresponding vowels in the corresponding age sample groups, can be obtained by solving the matrix constructed after pairwise comparison and scoring using the analytic hierarchy process based on the proportion of regional samples.

[0121] The new standard feature range was obtained by adjusting the expected feature range of corresponding consonants in the age sample group using the general consonant feature range and the regional sample proportion.

[0122] The new standard feature range can be calculated by directly weighting the expected feature range of the corresponding consonant in the corresponding age sample group by taking the general consonant feature range whose differences from the historical standard feature range of the same consonant in the same region exceed the preset difference constraint.

[0123] The weights assigned to the historical standard feature range of the same consonant in the same region that exceed the preset difference constraint of the general consonant feature range, and the expected feature range of the corresponding consonant in the corresponding age sample group, can be obtained by solving the matrix constructed after pairwise comparison and scoring using the analytic hierarchy process based on the regional sample proportion.

[0124] The beneficial effects of the above technical solution are as follows: By dividing new healthy population speech data samples by age and extracting vowels and auxiliary features to determine the first reference position interval and feature range, the unique acoustic features of different age groups in speech pronunciation can be accurately captured, providing targeted basic data for subsequent updates to the standard range; by comparing the first reference interval and range with historical standards and determining the expected interval and range based on whether the degree of difference exceeds preset constraints, it is ensured that new references can be adopted in a timely manner when pronunciation features change significantly to adapt to dynamic changes in speech features, while also taking into account historical standards through weighted averaging when changes are not significant; finally, by introducing regional language pronunciation differences to adjust the expected interval and range, dynamic and accurate updates to the standard range can be achieved, which helps to provide more accurate, reliable and realistic vowel and consonant standard references for the early identification of Parkinson's disease dysarthria in different age and regional scenarios.

[0125] The working principle of the above technical solution is as follows: First, the new samples are divided into multiple age sample groups according to age. Based on the vowel features (multiple formant frequencies) and consonant acoustic features of the extracted speech data samples in each age sample group, the first reference position interval for vowels and the first reference feature range for consonants are determined. Then, the first reference position interval and the first reference feature range are compared with the historical standard position interval and the historical standard feature range of the same age interval to obtain the degree of difference between the position interval and the degree of difference between the feature range. Next, when the degree of difference exceeds the preset difference constraint, the first reference position interval or the first reference feature range is taken as the expected position interval. Alternatively, the expected feature range can be determined; otherwise, the position interval and feature range of the first reference and the historical standard can be weighted and averaged to obtain the expected position interval and expected feature range. Then, the samples are divided into age groups according to region, the regional sample proportion is determined, and the general vowel position interval and general consonant feature range of the region are obtained. Finally, the expected position interval is adjusted by combining the general vowel position interval with the regional sample proportion, which has a difference degree exceeding the preset difference constraint, to obtain a new standard position interval. The expected feature range is adjusted by combining the general consonant feature range with the regional sample proportion, which has a difference degree exceeding the preset difference constraint, to obtain a new standard feature range, thus realizing the update of the standard range.

[0126] The process of extracting vowel-related parameters and speech fluency indicators includes:

[0127] For each vowel segment, a two-dimensional acoustic space is constructed based on the formant frequencies, and the distance between different vowels (such as ai, au, iu) is calculated to determine the area of ​​the vowel space;

[0128] In calculating the vowel space area, if the object to be identified has some missing vowel pronunciations, the interpolation method is used to supplement the standard position of the missing vowel based on the existing vowel acoustic features, and then the vowel space area is calculated. The interpolation weight is determined according to the acoustic correlation between the missing vowel and the existing vowel.

[0129] The clarity, stability, and duration of vowel pronunciation are calculated, and a vowel pronunciation index is obtained by weighted summation. The weights in the weighted summation are preset based on the sensitivity of different vowels to articulation disorders.

[0130] The effective speech segments are divided into time segments, and pauses in the speech signal are identified. The pauses are the periods when the signal amplitude is lower than a preset amplitude threshold. The duration of each pause is counted to obtain the speech pause duration distribution.

[0131] The number of written syllables corresponding to speech per unit time is calculated as the speech rate value, and the rate of change of speech rate in different time periods is calculated using the sliding window method as the speech rate change index.

[0132] Calculating the clarity of vowel pronunciation includes:

[0133] Obtain the preset cepstral features of vowels; the preset cepstral features include cepstral peak height, number of cepstral peaks, and cepstral peak position;

[0134] The deviation analysis is performed on the cepstral peak position of the vowel and the formant position of the corresponding standard vowel and the preset theoretical position of the corresponding vowel category, respectively, to obtain the first position deviation status and the second position deviation status.

[0135] When both the first position deviation condition and the second position deviation condition are less than the corresponding position deviation threshold, the larger value between the first position deviation condition and the second position deviation condition is selected as the reference position deviation.

[0136] Otherwise, the average of the first and second position deviation conditions shall be regarded as the reference position deviation;

[0137] The clarity of vowel pronunciation is calculated using cepstral peak height, cepstral peak number, and reference position deviation.

[0138] The preset cepstral feature is the peak correlation feature obtained by performing cepstral analysis on the vowel speech signal; cepstral analysis is to take the logarithm of the signal spectrum and then perform an inverse Fourier transform to obtain the cepstral.

[0139] The cepstral peak height is the amplitude value of the peak point in the cepstral spectrum, used to reflect the energy intensity of a specific frequency component of a vowel signal in the cepstral domain.

[0140] The number of cepstral peaks is the number of identifiable peaks in the cepstral spectrum, which can be determined by counting the number of peaks exceeding a preset threshold (e.g., twice the background noise). The cepstral peak position represents the frequency location of the peak in the cepstral domain, which can be obtained using a peak detection algorithm.

[0141] The standard vowel formant position is the frequency at which the formant is located when a vowel is pronounced, as determined through extensive research and experimentation and considered to be typical or standard.

[0142] Vowel categories are defined according to the International Phonetic Alphabet or linguistic classification standards (such as / a / , / i / , / u / , etc.).

[0143] The preset theoretical position is the average position of the cepstral peak under standard pronunciation conditions.

[0144] The first positional deviation is used to assess the physiological accuracy of vowel pronunciation. It is obtained by summing the squares of the deviations of the current vowel's cepstral peak heights F1 and F2 from the corresponding standard vowel formant positions (i.e., formant frequencies) and then taking the square root.

[0145] The second positional deviation condition is used to assess habitual deviations in vowel pronunciation. It is obtained by summing and taking the square root of the squares of the deviations of the current vowel's cepstral peak heights F1 and F2 from the corresponding preset theoretical positions (i.e., the average position of the cepstral peak).

[0146] The positional deviation threshold is the upper limit of the permissible deviation range determined based on statistical data from a large number of healthy speakers.

[0147] For example, the position deviation threshold for the first position deviation condition can be set to ±10% of the standard resonant frequency; the position deviation threshold for the second position deviation condition can be set to ±15% of the preset theoretical position.

[0148] The reference position deviation is determined by comparing the first position deviation condition, the second position deviation condition, and the corresponding position deviation threshold, and is used for the deviation condition in the sharpness calculation.

[0149] The formula for calculating the clarity of vowel pronunciation is as follows:

[0150]

[0151] In the formula, This indicates the clarity of the current vowel pronunciation; This represents the weight of the influence of cepstral peak condition on the clarity of vowel pronunciation, with a value of [value missing]. C represents the number of peaks in the current vowel pronunciation; Let represent the peak height of the i-th vowel sound, where i = 1, 2, ... C; This is expressed as the mean of the background noise; This represents the weight of the effect of positional offset on the clarity of vowel pronunciation, with values ​​ranging from [value missing]. ; This is expressed as a reference position deviation; This is represented as a preset position deviation constraint;

[0152] It should be noted that the preset position deviation constraint It is the positional deviation limit determined through statistical analysis of numerous vocalization experiments on healthy individuals.

[0153] The weights assigned to the cepstral peak condition and positional offset are obtained by solving the matrix constructed by pairwise comparison and scoring using the analytic hierarchy process.

[0154] The beneficial effects of the above technical solution are as follows: through multidimensional cepstral feature analysis, it can accurately capture subtle changes such as reduced vowel pronunciation energy, distorted formant structure and position shift caused by abnormal muscle control in Parkinson's disease patients, determine the clarity of vowel pronunciation, and provide an effective basis for subsequent calculation of vowel pronunciation index, thereby helping to improve the sensitivity of early identification.

[0155] The process of extracting sound roughness and pronunciation speed features includes:

[0156] Perform a short-time Fourier transform on the effective speech segment to obtain the spectrum, and calculate the energy ratio of harmonic components to noise components in the spectrum;

[0157] The degree of irregular fluctuation in the speech signal was analyzed, and the roughness of the sound was quantified by weighting the energy ratio and the degree of fluctuation. The weights were determined based on the statistics of clinical dysarthria samples.

[0158] The total number of syllables is counted based on the text sequence corresponding to the speech segment, and the average pronunciation speed (syllables / second) is calculated in combination with the total duration of the speech segment.

[0159] The transition duration between adjacent syllables during continuous pronunciation is analyzed, and the standard deviation of the transition duration is calculated as a characteristic of pronunciation speed stability. The average pronunciation speed and the speed stability characteristic are used together as characteristics related to pronunciation speed.

[0160] In the above embodiments, the vowel articulation index comprehensively considers key indicators such as clarity and stability of pronunciation through multi-dimensional weighted summation, highlighting the contribution of vowel features that are sensitive to articulation disorders. The speech fluency index is extracted from two dimensions: pause duration and speech rate variation, capturing common speech disfluency manifestations in patients with articulation disorders. The application of the sliding window method enables dynamic monitoring of speech rate variations. The harshness of sound is quantified through spectral analysis and fluctuation degree, accurately depicting changes in sound quality caused by abnormalities in the speech organs. The pronunciation speed feature takes into account both average speed and stability, comprehensively reflecting the fluency and coordination of pronunciation. Feature extraction covers the core performance dimensions of articulation disorders, providing rich and effective feature support for subsequent target feature screening and model recognition, and improving the comprehensiveness and accuracy of early identification.

[0161] When selecting target features, the selection process for vowel space area, vowel articulation index, speech pause duration, speech rate variation, voice roughness, and pronunciation speed includes:

[0162] The t-test method was used to calculate the significant differences in vowel space area, vowel articulation index, speech pause duration, speech rate variation, voice roughness, and speech speed between Parkinson's disease dysarthria samples and healthy population samples. Features with p-values ​​less than 0.05 were retained to obtain a preliminary feature set.

[0163] A feature importance evaluation model was constructed, with the accuracy of speech disorder recognition as the objective function. The importance score of each feature in the preliminary feature set was calculated using the random forest algorithm.

[0164] Sort the features by importance score from high to low, select the top N features, and combine them with time domain, frequency domain and nonlinear features to form the target features. N is preset according to the model complexity, and the preset range of N is 10-20.

[0165] When selecting the top N features by score, vowel space area, vowel articulation index, sound roughness, and articulation speed are mandatory features.

[0166] In the above embodiments, a feature importance evaluation model is constructed using the random forest algorithm, with recognition accuracy as the objective function. This objectively quantifies the contribution of each feature to the recognition result, avoiding feature bias caused by subjective experience-based selection. The top N features are selected according to their importance scores, ensuring high discriminative power of the target features while controlling the feature dimension through the range of N, thus avoiding model overfitting and balancing recognition accuracy with model efficiency. Core features such as vowel space area are set as mandatory features, ensuring that key recognition dimensions of articulation disorders are not overlooked. This solves the problem that core features may be lost due to score ranking in traditional feature selection. The entire selection process is progressive, achieving precise selection and optimized combination of features. The final target feature set has discriminative power, representativeness, and simplicity, providing a core guarantee for the efficient and accurate recognition of the model and improving the performance of early recognition of articulation disorders in Parkinson's disease.

[0167] The process of identifying high-risk individuals in the prodromal stage by combining characteristics such as vocal roughness and speech speed includes:

[0168] Based on a sample of patients with dysarthria in the prodromal stage of Parkinson's disease, high-risk threshold ranges for voice roughness (e.g., quantification value 0.6-0.8) and high-risk threshold ranges for speech rate (average speech rate less than 3 syllables / second, and speed stability standard deviation greater than 0.5) were determined.

[0169] The quantification value of the roughness of the voice and the characteristic value of the pronunciation speed of the subject to be identified are judged. If they both fall into the corresponding high-risk threshold range, and the identification results of other target features are suspected articulation disorders, then they are marked as high-risk groups in the prodromal stage.

[0170] For high-risk groups, the joint risk value of voice roughness and speech speed features is calculated. The joint risk value is obtained by weighted summation of the standardized values ​​of the two. The weights are determined based on the recall rate of high-risk group identification. When the joint risk value is greater than the preset risk threshold, a high-risk warning result is output.

[0171] Extract the temporal variation trends of voice roughness and pronunciation speed characteristics of high-risk groups, combine them with historical voice data to analyze the deterioration rate of the characteristics, and shorten the re-examination interval when the deterioration rate exceeds a preset rate threshold.

[0172] In the above embodiments, by using dual threshold judgments based on sound roughness and pronunciation speed characteristics, combined with the suspected identification results of other target features, preliminary accurate labeling of high-risk groups is achieved. The dual screening mechanism reduces the risk of misjudgment and improves the reliability of labeling. The calculation of the joint risk value comprehensively considers the risk contribution of the two core features through standardized processing and weighted summation. The weights are determined based on recall optimization, ensuring the sensitivity of high-risk group identification and avoiding the problem of missed judgment caused by single feature judgment. The analysis of the deterioration rate of high-risk group characteristics and the dynamic adjustment of the re-examination interval enable dynamic monitoring and precise management of high-risk groups, providing a time window for early intervention and better meeting the clinical needs for monitoring and intervention of prodromal diseases, significantly improving the clinical application value of early identification.

[0173] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for early identification of dysarthria in Parkinson's disease based on voiceprint features, characterized in that, Includes the following steps: Acquire the raw speech data of the object to be identified and preprocess it; Speech analysis and speech recognition are performed on the preprocessed speech data to determine the effective speech segments and semantic information corresponding to the speech content. At the same time, vowel segments and consonant segments are extracted from the speech data, and pronunciation deviations of consonants and vowels are identified. Speech processing is performed on valid speech segments to extract an initial feature set related to the voiceprint within the segment. The initial feature set includes the time-domain features, frequency-domain features, nonlinear features, vowel-related parameters, and language fluency index of the voiceprint. Feature selection is performed on the initial feature set to screen out target features that have significant discriminative power in identifying articulation disorders in Parkinson's disease. The target features include vowel space area, vowel articulation index, speech pause duration, speech rate variation, voice roughness, and speech speed. The target features are input into a pre-trained Parkinson's disease dysarthria recognition model. The model calculates the dysarthria recognition results of the subject to be identified, thus completing the identification of high-risk groups in the prodromal stage and achieving early identification. The process of extracting vowel-related parameters and language fluency indicators includes: For each vowel segment, a two-dimensional acoustic space is constructed based on the formant frequencies, the distance between different vowels is calculated, and the area of ​​the vowel space is determined. In calculating the vowel space area, if the object to be identified has some missing vowel pronunciations, the interpolation method is used to supplement the standard position of the missing vowel based on the existing vowel acoustic features, and then the vowel space area is calculated. The interpolation weight is determined according to the acoustic correlation between the missing vowel and the existing vowel. The clarity, stability, and duration of vowel pronunciation are calculated, and a vowel pronunciation index is obtained by weighted summation. The effective speech segments are divided into time segments, pauses in the speech signal are identified, the duration of each pause is counted, and the speech pause duration distribution is obtained. The number of written syllables corresponding to speech per unit time is calculated as the speech rate value, and the rate of change of speech rate in different time periods is calculated using the sliding window method as the speech rate change index. The calculation of vowel pronunciation clarity includes: Obtain the preset cepstral features of the vowel; the preset cepstral features include cepstral peak height, cepstral peak number, and cepstral peak position; The cepstral peak position of the vowel is analyzed to compare it with the corresponding standard vowel formant position and the preset theoretical position of the vowel category, respectively, to obtain the first position deviation status and the second position deviation status. When both the first position deviation condition and the second position deviation condition are less than the corresponding position deviation threshold, the larger value between the first position deviation condition and the second position deviation condition is selected as the reference position deviation; Otherwise, the average of the first position deviation and the second position deviation shall be regarded as the reference position deviation; The intelligibility of vowel pronunciation is calculated using the cepstral peak height, the number of cepstral peaks, and the reference position deviation. The formula for calculating the intelligibility of vowel pronunciation is as follows: In the formula, This indicates the clarity of the current vowel pronunciation; This represents the weight of the influence of cepstral peak condition on the clarity of vowel pronunciation, with a value of [value missing]. C represents the number of peaks in the current vowel pronunciation; Let represent the peak height of the i-th vowel sound, where i = 1, 2, ... C; This is expressed as the mean of the background noise; This represents the weight of the effect of positional offset on the clarity of vowel pronunciation, with values ​​ranging from [value missing]. ; This is expressed as a reference position deviation; This is represented as a preset position deviation constraint.

2. The method for early identification of dysarthria in Parkinson's disease based on voiceprint features as described in claim 1, characterized in that, The process of extracting vowel and consonant segments from speech data and identifying pronunciation deviations between consonants and vowels includes: Based on the text sequence obtained from speech recognition, the pronunciation periods of corresponding vowels and consonants in the speech data are determined, and vowel segments and consonant segments are extracted. Calculate the formant frequencies of vowel segments, establish the acoustic spatial distribution of vowels, and determine the standard position range of each vowel in acoustic space; Consonant acoustic features are extracted from consonant segments, including plosive duration, friction intensity, and voicing degree, and the standard feature range for each consonant is determined. In determining the standard position interval of vowels and the standard feature range of consonants, based on the speech data of healthy people, a 95% confidence interval is obtained through statistical analysis, which serves as the standard position interval and standard feature range. The standard range is updated periodically based on new speech data samples from healthy people, and the number of speech data samples from healthy people is not less than 1,000. The deviation values ​​of the actual acoustic spatial position of the vowel segment to be identified from the standard position range are compared with the deviation values ​​of the actual acoustic features of the consonant segment from the standard feature range. When the deviation value is greater than the preset deviation threshold, it is determined that there is a pronunciation deviation, and the deviation type and degree are recorded.

3. The method for early identification of dysarthria in Parkinson's disease based on voiceprint features as described in claim 2, characterized in that, The periodic updating of the standard range based on new voice data samples from healthy individuals includes: The voice data samples of the new healthy population are divided according to age to obtain multiple age sample groups; the multiple age sample groups are multiple sample groups composed of voice data samples of healthy people belonging to the same age range. Vowel features and auxiliary features were extracted from the speech data samples of each age group to obtain the analytical vowel features and analytical auxiliary features; Based on the analyzed vowel features, the first reference position interval of vowels in the age sample group is determined; Based on the aforementioned analytical auxiliary features, a first reference feature range for consonants in the corresponding age sample group is determined. The degree of difference between the first reference position interval and the historical standard position interval of the same vowel in the same age interval is obtained by comparing the differences between the position intervals. The difference between the first reference feature range and the historical standard feature range of the same consonant in the same age range is compared to obtain the degree of difference in the feature range. When the degree of difference in the position interval or the degree of difference in the feature range exceeds the corresponding preset difference constraint, the first reference position interval or the first reference feature range is regarded as the expected position interval of the corresponding vowel or the expected feature range of the corresponding consonant in the current age sample group. When the degree of difference in the position interval or the degree of difference in the feature range does not exceed the corresponding preset difference constraint, the expected position interval of the corresponding vowel in the current age sample group is obtained by weighted averaging the first reference position interval and the historical standard position interval, or the expected feature range of the corresponding consonant in the current age sample group is obtained by weighted averaging the first reference feature range and the historical standard feature range. The sample statistics of each age group were carried out according to different regions to determine the sample proportion of multiple regions. Obtain the common vowel position range and common consonant feature range of the region; For all general vowel position intervals whose differences from the historical standard position intervals of the same vowel in the same region exceed the preset difference constraint, the expected position intervals of the corresponding vowels in the age sample group are adjusted in combination with the regional sample proportion to obtain new standard position intervals. The expected feature range of the corresponding consonant in the age sample group is adjusted by taking into account the regional sample proportion and the degree of difference between the historical standard feature range of all consonants in the same region and the general consonant feature range that exceeds the preset difference constraint, so as to obtain a new standard feature range.

4. The method for early identification of dysarthria in Parkinson's disease based on voiceprint features as described in claim 1, characterized in that, The process of extracting sound roughness and pronunciation speed features includes: Perform a short-time Fourier transform on the effective speech segment to obtain the spectrum, and calculate the energy ratio of harmonic components to noise components in the spectrum; The degree of irregular fluctuation in the speech signal is analyzed, and the quantification value of the sound roughness is obtained by weighting the energy ratio and the degree of fluctuation. The average pronunciation speed is calculated by counting the total number of syllables in the text sequence corresponding to the speech segment and combining it with the total duration of the speech segment. The transition duration between adjacent syllables during continuous pronunciation is analyzed, and the standard deviation of the transition duration is calculated as a characteristic of pronunciation speed stability. The average pronunciation speed and the speed stability characteristic are used together as characteristics related to pronunciation speed.

5. The method for early identification of dysarthria in Parkinson's disease based on voiceprint features as described in claim 1, characterized in that, When selecting target features, the selection process for vowel space area, vowel articulation index, speech pause duration, speech rate variation, voice roughness, and pronunciation speed includes: The t-test method was used to calculate the significant differences in vowel space area, vowel articulation index, speech pause duration, speech rate variation, voice roughness, and speech speed between Parkinson's disease dysarthria samples and healthy population samples. Features with p-values ​​less than 0.05 were retained to obtain a preliminary feature set. A feature importance evaluation model was constructed, with the accuracy of speech disorder recognition as the objective function. The importance score of each feature in the preliminary feature set was calculated using the random forest algorithm. Sort the features by importance score from high to low, select the top N features, and combine them with time domain, frequency domain and nonlinear features to form the target features. N is preset according to the model complexity, and the preset range of N is 10-20. When selecting the top N features by score, vowel space area, vowel articulation index, sound roughness, and articulation speed are mandatory features.

6. The method for early identification of dysarthria in Parkinson's disease based on voiceprint features as described in claim 1, characterized in that, The process of identifying high-risk individuals in the prodromal stage by combining characteristics such as vocal roughness and speech speed includes: Based on samples of patients with dysarthria in the preclinical stage of Parkinson's disease, high-risk threshold ranges for voice roughness and speech speed were determined. The quantification value of the roughness of the voice and the characteristic value of the pronunciation speed of the subject to be identified are judged. If they both fall into the corresponding high-risk threshold range, and the identification results of other target features are suspected articulation disorders, then they are marked as high-risk groups in the prodromal stage. For high-risk groups, a joint risk value is calculated based on the roughness of the voice and the speed of speech. The joint risk value is obtained by weighted summation of the standardized values ​​of the two. When the joint risk value is greater than a preset risk threshold, a high-risk warning result is output.

7. The method for early identification of dysarthria in Parkinson's disease based on voiceprint features as described in claim 6, characterized in that, After outputting the high-risk warning results, it also includes: Extract the temporal variation trends of voice roughness and pronunciation speed characteristics of high-risk groups, combine them with historical voice data to analyze the deterioration rate of the characteristics, and shorten the re-examination interval when the deterioration rate exceeds a preset rate threshold.

8. The method for early identification of dysarthria in Parkinson's disease based on voiceprint features as described in claim 1, characterized in that, The acquired raw speech data must include preset speech disorder sensitive sentences, which cover high-frequency vowels, consonant combinations, and content with different speech rates to ensure the effective extraction of vowel segments and consonant segments and the accurate calculation of related parameters.

Images