Method for Automatic Classification of Autism Spectrum Disorder Using Speech Features

Abstract

The present invention relates to an automatic classification method for autism spectrum disorder that classifies autism spectrum disorder through a machine learning classifier by reflecting characteristics related to voice quality, pitch, speech rate, and voice spectrum, which are voice characteristics of autistic children, for the purpose of providing objective and consistent discrimination results based on extracted voice features. The automatic classification method for autism spectrum disorder using voice features according to the present invention has the advantage of not only increasing the accuracy of the discrimination model but also enabling intuitive interpretation of the results by reflecting clinical knowledge into the automatic discrimination model using selected features corresponding to voice characteristics in which a person with autism spectrum disorder exhibits patterns different from a control group.

Description

Technology Field

[0001] The present invention relates to an automatic classification method for autism spectrum disorder, and more specifically, to an automatic classification method for autism spectrum disorder that classifies autism spectrum disorder through a machine learning classifier by reflecting voice quality, pitch, speech rate, and voice spectrum-related characteristics, which are voice characteristics of autistic children, for the purpose of providing objective and consistent discrimination results based on extracted voice features. Background Technology

[0003] According to data from the Korean Society of Pediatric Neurology, Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that typically manifests within the first three years of life. It is characterized by deficits in social and language abilities, accompanied by repetitive interests or behaviors. Autism Spectrum Disorder is known to be associated with the failure of normal neural connections within the cerebral cortex, an area composed of nerve cells called neurons. In other words, as neural circuits in the cerebral cortex develop less than normal growth patterns, a lack of important connections between cortical regions leads to the manifestation of symptoms involving deficits in language, social skills, and behavior.

[0004] Children with autism spectrum disorders exhibit behavioral characteristics such as the repetitive use of specific objects, as well as repetitive language usage characteristics like echolalia, neologisms, and the use of personal phrases. The majority of individuals with autism spectrum disorder have language impairments, ranging from complete inability to speak to language delays, poor comprehension of speech, unnatural speech rhythm, or a lack of pragmatic skills that prevents them from understanding connotative expressions or metaphors. Therefore, diagnosing autism spectrum disorders requires neurological, cognitive, and language assessments, as well as hearing tests to confirm the presence of hearing impairment.

[0005] Korean Public Patent No. 2023-0046199 relates to [Deep Learning-based Device and Method for Predicting Autism Spectrum Disorder in Infants and Toddlers], disclosing a technology regarding a deep learning-based device and method for predicting autism spectrum disorder in infants and toddlers using autoencoder feature representations, which can screen for autism spectrum disorder from the voice of infants and toddlers using autoencoder feature representations. However, the above technology has a problem in that the extracted features are embeddings of a hidden layer containing general information about the voice, and thus include features unrelated to autism spectrum disorder, such as voice disorders. Furthermore, considering that automatic evaluation results in clinical settings must be interpretable, the above technology has limitations in that it is difficult to directly apply it because it is difficult to verify which features were effectively operated during the evaluation process. Prior art literature

[0007] Republic of Korea Published Patent No. 2023-0046199 The problem to be solved

[0008] The present invention proposes a knowledge-driven voice feature set capable of reflecting the voice characteristics of children with autism spectrum disorder, and aims to provide an automatic classification method for autism spectrum disorder using voice features that exhibits superior classification performance compared to the comprehensive feature set of the prior art, with a high degree of suitability for classifying autism spectrum disorder. means of solving the problem

[0010] The present invention provides an automatic classification method for autism spectrum disorder using speech features, wherein the method comprises: a step of acquiring and storing a meaningful Korean utterance from a subject to test by a computer including a processor and memory, and retrieving the stored Korean utterance; a step of extracting pitch, timbre, speech rate, and speech spectrum features from the retrieved Korean utterance by using an analysis toolkit by a computer including a processor and memory; a step of inputting the extracted pitch, timbre, speech rate, and speech spectrum features into a trained machine learning automatic classifier by a computer including a processor and memory; and a step of outputting whether the subject to test has autism spectrum disorder from the trained machine learning automatic classifier by a computer including a processor and memory, wherein the trained machine learning automatic classifier is a Support Vector Machine (SVM) that classifies whether there is autism spectrum disorder based on the input pitch, timbre, speech rate, and speech spectrum features.

[0011] The present invention also comprises a support vector machine that classifies speech as autism spectrum disorder, wherein the trained machine learning automatic classifier has passed KDSF method training, and the KDSF method comprises: a step of acquiring speech data composed of articulations including all Korean consonant and vowel phonemes from a plurality of speakers including a person with Autism Spectrum Disorder (ASD) by a computer including a processor and memory; a step of preprocessing speech data by a computer including a processor and memory, after removing personal information from the speech data, segmenting the speech into utterance units, executing phoneme transcription including spelling transcription, and tagging specific utterances; a step of extracting pitch, timbre, speech rate, and speech spectrum features from the preprocessed speech data using an analysis toolkit by a computer including a processor and memory; and a step of inputting the pitch, timbre, speech rate, and speech spectrum features to the support vector machine by a computer including a processor and memory. A method for automatically classifying autism spectrum disorder is provided, comprising: a step of comparing an autism spectrum disorder classification result classified and output by a support vector machine with an autism spectrum disorder classification result input and output after eGeMAPS feature extraction, by a computer including a processor and memory; and a step of acknowledging that it is a trained machine learning automatic classifier when the difference between the two autism spectrum disorder classification results in the upper bound comparison step is within a predetermined error range.

[0012] The present invention also provides an automatic classification method for autism spectrum disorder, wherein the analysis toolkit is librosa for extracting a voice spectrum and Parselmouth for extracting pitch, voice quality, and speech rate.

[0013] The present invention also provides an automatic classification method for autism spectrum disorder, wherein the trained machine learning classifier inputs pitch, timbre, speech rate, and speech spectrum features extracted from voice files of at least 9 control group (TD: typically developing) speakers and at least 60 speakers with autism spectrum disorder into the support vector machine, and has data classifying the voice into a plurality of autism spectrum disorders.

[0014] The present invention also provides an automatic classification method for autism spectrum disorder, wherein the pitch characteristics of the voice include, as sub-features, the mean, standard deviation, minimum value, maximum value, median, 25th percentile, and 75th percentile of the utterance; the voice quality characteristics of the voice include, as sub-features, jitter, shimmer, HNR (Harmonic-to-Noise Ratio), the number of voice breaks, and the ratio of voice breaks; the speech speed characteristics of the voice include, as sub-features, the total length of the utterance, the length of the pause interval within the utterance, the length of the pause-excluded interval within the utterance, speech rate, articulation rate, average syllable length, the number of pauses, and the ratio of speech sound intervals to total utterance intervals; and the voice spectrum characteristics of the voice include log energy and 1st to 12th order MFCC (Mel-frequency cepstral coefficients) values.

[0015] The present invention also provides a computer-readable storage medium for storing a computer program for automatically classifying autism spectrum disorder, wherein the storage medium comprises: a step of acquiring and storing a meaningful Korean utterance from a test subject and retrieving the stored Korean utterance; a code portion programmed to extract pitch, phonological quality, speech rate, and speech spectrum features from the retrieved Korean utterance using an analysis toolkit by a computer including a processor and memory; a code portion programmed to input the extracted pitch, phonological quality, speech rate, and speech spectrum features into a trained machine learning automatic classifier; and a code portion programmed to output whether the test subject has autism spectrum disorder from the trained machine learning automatic classifier, wherein the trained machine learning automatic classifier is a Support Vector Machine (SVM) that classifies whether there is autism spectrum disorder based on the input pitch, phonological quality, speech rate, and speech spectrum features.

[0016] The present invention also comprises a support vector machine that classifies speech as autism spectrum disorder, wherein the trained machine learning automatic classifier has passed KDSF method training, and the KDSF method comprises: a step of acquiring speech data composed of articulations including all Korean consonant and vowel phonemes from a plurality of speakers including a person with Autism Spectrum Disorder (ASD) by a computer including a processor and memory; a step of preprocessing speech data by a computer including a processor and memory, after removing personal information from the speech data, segmenting the speech into utterance units, executing phoneme transcription including spelling transcription, and tagging specific utterances; a step of extracting pitch, timbre, speech rate, and speech spectrum features from the preprocessed speech data using an analysis toolkit by a computer including a processor and memory; and a step of inputting the pitch, timbre, speech rate, and speech spectrum features to the support vector machine by a computer including a processor and memory. A computer-readable storage medium is provided, comprising: a step of comparing an autism spectrum disorder classification result classified and output by a support vector machine with an autism spectrum disorder classification result input and output after eGeMAPS feature extraction, by a computer including a processor and memory; and a step of acknowledging that it is a trained machine learning automatic classifier when the difference between the two autism spectrum disorder classification results in the upper bound comparison step is within a predetermined error range.

[0017] The present invention also provides a computer-readable storage medium in which the analysis toolkit is librosa for extracting a voice spectrum and Parselmouth for extracting pitch, voice quality, and speech rate.

[0018] The present invention also provides a computer-readable storage medium having data in which the trained machine learning classifier inputs pitch, timbre, speech rate, and speech spectrum features extracted from voice files of at least 9 control (TD: typically developing) speakers and at least 60 speakers with autism spectrum disorder into the support vector machine to classify the voice into a plurality of autism spectrum disorders.

[0019] The present invention also provides a computer-readable storage medium wherein the pitch characteristics of the voice include, as sub-features, the mean, standard deviation, minimum value, maximum value, median, 25th percentile, and 75th percentile of the utterance; the voice quality characteristics of the voice include, as sub-features, jitter, shimmer, HNR (Harmonic-to-Noise Ratio), the number of voice breaks, and the ratio of voice breaks; the voice velocity characteristics of the voice include, as sub-features, the total length of the utterance, the length of the pause interval within the utterance, the length of the pause-excluded interval within the utterance, speech rate, articulation rate, average syllable length, the number of pauses, and the ratio of the speech sound interval to the total utterance interval; and the voice spectrum characteristics of the voice include log energy and 1st to 12th order MFCC (Mel-frequency cepstral coefficients) values. Effects of the invention

[0021] The automatic classification method for autism spectrum disorder using voice features of the present invention has the advantage of not only improving the accuracy of the discrimination model but also enabling intuitive interpretation of the results by reflecting clinical knowledge in the automatic discrimination model through the selection and use of features corresponding to voice characteristics in which individuals with autism spectrum disorder exhibit different patterns from the control group. Brief explanation of the drawing

[0023] FIG. 1 is a conceptual diagram of a method for automatically classifying autism spectrum disorder by inputting features extracted from speech into a machine learning classifier according to one embodiment of the present invention. Specific details for implementing the invention

[0024] Various aspects are disclosed with reference to the drawings. For illustrative purposes, the following description discloses numerous specific details to aid in a general understanding of one or more aspects. However, it will be recognized that these aspects may be practiced without each specific detail. The following description and the attached drawings describe specific exemplary aspects of one or more aspects in detail. However, these aspects are exemplary, and some of the various methods may be used in the principles of the various aspects, and the description is intended to include all such aspects and their equivalents.

[0025] Various aspects and features will be presented by a system that may include multiple devices, modules, etc. It should also be understood and recognized that various systems may include additional devices, parts, components, etc. and / or may not include all of the devices, parts, components, etc. discussed in relation to the drawings.

[0026] Terms such as “embodiment,” “example,” “aspect,” “example,” etc., as used herein, shall not be construed as implying that any described aspect or design is superior or advantageous to other aspects or designs. Terms such as “system,” “server,” and “terminal” used below generally refer to a computer-related entity and may refer, for example, to hardware, a combination of hardware and software, or software.

[0027] Furthermore, the term "or" is intended to mean an implicit "or" rather than an exclusive "or." That is, unless otherwise specified or evident from the context, "X uses A or B" is intended to mean one of the natural implicit substitutions. In other words, if X uses A; if X uses B; or if X uses both A and B, "X uses A or B" may apply to either of these cases. Additionally, the term "and / or" as used herein should be understood to refer to and include all possible combinations of one or more of the enumerated related items.

[0028] Furthermore, the terms “comprising” and / or “comprising” should be understood to mean that the relevant feature, step, action, module, and / or component is present, but not to exclude the presence or addition of one or more other features, steps, actions, modules, components, and / or groups thereof. Additionally, while terms such as “first,” “second,” etc., may be used in this specification to describe various components, these components are not limited by such terms. That is, such terms are used merely to distinguish between two or more components and should not be interpreted as implying order or priority. Furthermore, unless otherwise specified or clearly evident from the context to indicate a singular form, the singular in this specification and claims should generally be interpreted to mean “one or more.”

[0029] Autism spectrum disorder is characterized by deficits in social communication and restricted, stereotyped behaviors and interests. Diagnosis of autism spectrum disorder is conducted by observing the child's behavior in a clinical setting; however, since this diagnostic process is time-consuming, it places a burden not only on clinicians but also on the child and their guardian. Because children with autism spectrum disorder exhibit distinctive vocal characteristics, standardized diagnostic tools also include items that evaluate the child's speech and language. Therefore, this invention is a response to the need to reduce costs, such as time and effort, in the diagnostic process by automatically differentiating autism spectrum disorder based on the child's voice to assist in diagnosis.

[0030] A proposed technique for the automatic discrimination of autistic children involves a methodology that extracts a feature set of overall speech characteristics, selects features suitable for the data being used, and inputs them into a machine learning-based classifier (Beccaria et al. (2022). Extraction and Classification of Acoustic Features from Italian Speaking Children with Autism Spectrum Disorders. In Proceedings of RaPID-4 @ LREC 2022, 22-30). This methodology has a limitation in that it is difficult to apply if the data being used changes, as it selects features that are valid only for specific datasets. Embodiments of the present invention will be described below with reference to the attached drawings.

[0031] FIG. 1 is a conceptual diagram of a method for automatically classifying autism spectrum disorder by inputting features extracted from speech into a machine learning classifier according to an embodiment of the present invention. The present invention is a method for automatically classifying autism spectrum disorder using an automatic classifier comprising a processor and memory, comprising the steps of: extracting, respectively, a plurality of voice quality features, pitch features, speech rate features, and voice spectrum-related features from an input speech using an analysis toolkit; and inputting the voice quality features, pitch features, speech rate features, and voice spectrum-related features into a trained machine learning classifier.

[0032] In one embodiment of the present invention, an automatic classification method for autism spectrum disorder using speech features comprises: a step of acquiring and storing a meaningful Korean utterance from a subject and retrieving the stored Korean utterance by a computer including a processor and memory; a step of extracting pitch, timbre, speech rate, and speech spectrum features from the retrieved Korean utterance using an analysis toolkit by a computer including a processor and memory; a step of inputting the extracted pitch, timbre, speech rate, and speech spectrum features into a trained machine learning automatic classifier by a computer including a processor and memory; and a step of outputting whether the subject has autism spectrum disorder from the trained machine learning automatic classifier by a computer including a processor and memory. In one embodiment of the present invention, the trained machine learning automatic classifier is a Support Vector Machine (SVM) that classifies whether there is autism spectrum disorder based on the input pitch, timbre, speech rate, and speech spectrum features.

[0033] In one embodiment of the present invention, the trained machine learning automatic classifier is a support vector machine that classifies speech as autism spectrum disorder and has passed KDSF method training, wherein the KDSF method comprises: a step of acquiring speech data composed of articulations including all Korean consonant and vowel phonemes from a plurality of speakers including a person with autism spectrum disorder (ASD) by a computer including a processor and memory; a step of preprocessing speech data by a computer including a processor and memory, after removing personal information from the speech data, segmenting the speech into utterance units, executing phoneme transcription including spelling transcription, and tagging specific utterances; a step of extracting pitch, timbre, speech rate, and speech spectrum features from the preprocessed speech data using an analysis toolkit by a computer including a processor and memory; and a step of inputting the pitch, timbre, speech rate, and speech spectrum features to the support vector machine by a computer including a processor and memory. The method includes a step of comparing an autism spectrum disorder classification result classified and output by the support vector machine with an autism spectrum disorder classification result input and output after eGeMAPS feature extraction, by a computer including a processor and memory; and a step of acknowledging that it is a trained machine learning automatic classifier when the difference between the two autism spectrum disorder classification results in the upper bound comparison step is within a predetermined error range.

[0034] In one embodiment of the present invention, the trained machine learning classifier has data in which the voice is classified into a plurality of autism spectrum disorders by inputting pitch, voice quality, speech rate, and voice spectrum features extracted from voice files of at least 9 control group (TD: typically developing) speakers and at least 60 speakers with autism spectrum disorder into the support vector machine.

[0035] In one embodiment of the present invention, the voice analysis toolkit for extracting features using the acquired voice file primarily utilizes the following types of open source code. In one embodiment of the present invention, the voice file is analyzed through an input program using Python or the like that includes the voice analysis toolkit.

[0036] - Speech spectrum extraction: librosa 0.8.1

[0037] - Sound quality feature extraction: Parselmouth 0.4.0

[0038] - Ignition velocity feature extraction: Parselmouth 0.4.0

[0039] - Pitch Feature Extraction: Parselmouth 0.4.0

[0041] In one embodiment of the present invention, the analysis toolkit can be selected from librosa, which extracts a speech spectrum, and Parselmouth, which extracts pitch, voice quality, and speech rate.

[0042] In one embodiment of the present invention, the extracted voice quality features include jitter representing frequency variation, shimmer representing amplitude variation, HNR representing the degree of noise in the voice, the number of voice breaks, and the ratio of voice breaks. In one embodiment of the present invention, among the extracted voice features, the voice velocity includes the total length of the voice, the length of the pause section within the voice, the length of the pause-excluding section within the voice, speech speed, articulation speed, average syllable length, the number of pauses, and the ratio of the voice sound section to the total voice section; and among the extracted voice features, the pitch includes the mean value, standard deviation, minimum value, maximum value, median, 25th percentile, and 75th percentile of the voice.

[0043] In one embodiment of the present invention, regarding features related to speech rate, patients with autism spectrum disorder have a slower speech rate than non-disabled patients, longer speech and syllable lengths, longer pauses, and more frequent pauses; thus, the total length of the speech, speech length (a section excluding pauses from the total length), speech rate (a value obtained by dividing the total number of syllables by the total length), articulation rate (a value obtained by dividing the total number of syllables by the speech length), pause length (a pause section), number of pauses (a pause section), average syllable length (a speech length divided by the total number of syllables), and the ratio of speech length to total length are considered as features. In one embodiment of the present invention, the extracted speech spectrum is log energy and 1st to 12th order MFCC (Mel-frequency cepstral coefficients) values. In one embodiment of the present invention, the pitch characteristics of the voice include, as sub-features, the mean, standard deviation, minimum value, maximum value, median, 25th percentile, and 75th percentile of the utterance; the voice quality characteristics of the voice include, as sub-features, jitter, shimmer, HNR (Harmonic-to-Noise Ratio), the number of voice breaks, and the ratio of voice breaks; the voice velocity characteristics of the voice include, as sub-features, the total length of the utterance, the length of the pause interval within the utterance, the length of the pause-excluded interval within the utterance, speech rate, articulation rate, average syllable length, the number of pauses, and the ratio of the speech sound interval to the total utterance interval; and the voice spectrum characteristics of the voice include log energy and 1st to 12th order MFCC (Mel-frequency cepstral coefficients) values.

[0044] The present invention relates to a computer-readable storage medium for storing a computer program for automatically classifying autism spectrum disorder, wherein the storage medium comprises: a programmed code portion for acquiring and storing meaningful Korean utterances from a test subject and retrieving the stored Korean utterances; a code portion programmed to extract pitch, timbre, speech rate, and speech spectrum features from the retrieved Korean utterances using an analysis toolkit; a code portion programmed to input the extracted pitch, timbre, speech rate, and speech spectrum features into a trained machine learning automatic classifier; and a code portion programmed to output whether the test subject has autism spectrum disorder from the trained machine learning automatic classifier, wherein the trained machine learning automatic classifier is a Support Vector Machine (SVM) that classifies whether there is autism spectrum disorder based on the input pitch, timbre, speech rate, and speech spectrum features, and is operable on a computer. In one embodiment of the present invention, the computer may acquire the sentence utterances of the test subject speaker as voice files through methods such as recording.

[0045] In one embodiment of the present invention, the trained machine learning automatic classifier is a support vector machine that classifies speech as autism spectrum disorder and has passed KDSF method training, wherein the KDSF method comprises: a step of acquiring speech data composed of articulations including all Korean consonant and vowel phonemes from a plurality of speakers including a person with autism spectrum disorder (ASD) by a computer including a processor and memory; a step of preprocessing speech data by a computer including a processor and memory, after removing personal information from the speech data, segmenting the speech into utterance units, executing phoneme transcription including spelling transcription, and tagging specific utterances; a step of extracting pitch, timbre, speech rate, and speech spectrum features from the preprocessed speech data using an analysis toolkit by a computer including a processor and memory; and a step of inputting the pitch, timbre, speech rate, and speech spectrum features to the support vector machine by a computer including a processor and memory. The method may include a step of comparing an autism spectrum disorder classification result classified and output by the support vector machine with an autism spectrum disorder classification result input and output after eGeMAPS feature extraction, by a computer including a processor and memory; and a step of acknowledging that it is a trained machine learning automatic classifier when the difference between the two autism spectrum disorder classification results in the upper bound comparison step is within a predetermined error range.

[0046] In one embodiment of the present invention, the analysis toolkit is librosa for extracting speech spectrum and Parselmouth for extracting pitch, timbre, and speech rate, and the trained machine learning classifier inputs pitch, timbre, speech rate, and speech spectrum features extracted from speech files of at least 9 control (TD: typically developing) speakers and at least 60 speakers with autism spectrum disorder into the support vector machine to provide data classifying speech into a plurality of autism spectrum disorders. In one embodiment of the present invention, the pitch characteristics of the voice include, as sub-features, the mean, standard deviation, minimum value, maximum value, median, 25th percentile, and 75th percentile of the utterance; the voice quality characteristics of the voice include, as sub-features, jitter, shimmer, HNR (Harmonic-to-Noise Ratio), the number of voice breaks, and the ratio of voice breaks; the voice velocity characteristics of the voice include, as sub-features, the total length of the utterance, the length of the pause interval within the utterance, the length of the pause-excluded interval within the utterance, speech rate, articulation rate, average syllable length, the number of pauses, and the ratio of the speech sound interval to the total utterance interval; and the voice spectrum characteristics of the voice include log energy and 1st to 12th order MFCC (Mel-frequency cepstral coefficients) values.

[0047] According to software implementation, embodiments such as the procedures and functions described herein may be implemented in separate software modules. Each of said software modules may perform one or more functions and operations described herein. Software code may be implemented as a software application written in a suitable programming language. The software code may be stored in a management server and / or database and executed by an application.

[0048] Meanwhile, the various embodiments presented herein may be implemented as methods, devices, or articles manufactured using standard programming and / or engineering techniques. The term “article manufactured” includes a computer program, carrier, or medium accessible from any computer-readable device. For example, computer-readable media include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, magnetic strips, etc.), optical discs (e.g., CDs, DVDs, etc.), smart cards, and flash memory devices (e.g., EEPROMs, cards, sticks, key drives, etc.). Additionally, the various storage media presented herein include one or more devices and / or other machine-readable media for storing information. The term “machine-readable media” includes, but is not limited to, wireless channels and various other media capable of storing, holding, and / or transmitting command(s) and / or data.

[0049] The description of the presented embodiments is provided so that any person skilled in the art may use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments without departing from the scope of the present invention. Thus, the present invention is not limited to the embodiments presented herein, but should be interpreted in the broadest possible scope consistent with the principles and novel features presented herein.

Claims

Claim 1 A method for automatic classification of autism spectrum disorder using speech features, wherein the method comprises: a step of acquiring and storing meaningful Korean utterances from a test subject and retrieving the stored Korean utterances by means of a computer including a processor and memory; a step of extracting pitch, timbre, speech rate, and speech spectrum features from the retrieved Korean utterances using an analysis toolkit by means of a computer including a processor and memory; and a step of inputting the extracted pitch, timbre, speech rate, and speech spectrum features into a trained machine learning automatic classifier by means of a computer including a processor and memory. The method includes a step of outputting whether the subject has an autism spectrum disorder from the trained machine learning automatic classifier by a computer including a processor and memory, wherein the trained machine learning automatic classifier is a Support Vector Machine (SVM) that classifies whether there is an autism spectrum disorder based on the input pitch, phonological quality, speech rate, and speech spectrum features, and wherein the trained machine learning automatic classifier is a Support Vector Machine that classifies speech as autism spectrum disorder and has passed KDSF method training, and the KDSF method comprises: a step of acquiring speech data composed of articulations including all Korean consonant and vowel phonemes from a plurality of speakers including a person with autism spectrum disorder (ASD) by a computer including a processor and memory; a step of preprocessing speech data by removing personal information from the speech data, segmenting the speech into speech units, executing phoneme transcription including spelling transcription, and tagging specific speeches by a computer including a processor and memory; and from the preprocessed speech data by using an analysis toolkit by a computer including a processor and memory Step of extracting pitch, timbre, speech rate, and speech spectrum features;A method for automatically classifying autism spectrum disorder, comprising: a step of inputting the pitch, sound quality, speech rate, and speech spectrum features to a support vector machine by means of a computer including a processor and memory; a step of comparing an autism spectrum disorder classification result classified and output by the support vector machine with an autism spectrum disorder classification result input and output after eGeMAPS feature extraction by means of a computer including a processor and memory; and a step of acknowledging that it is a trained machine learning automatic classifier when the difference between the two autism spectrum disorder classification results in the upper bound comparison step is within a predetermined error range. Claim 2 delete Claim 3 In claim 1, the analysis toolkit is librosa for extracting a voice spectrum and Parselmouth for extracting pitch, voice quality, and speech rate, an automatic classification method for autism spectrum disorder. Claim 4 A method for automatically classifying autism spectrum disorder according to claim 1, wherein the trained machine learning automatic classifier inputs pitch, timbre, speech rate, and speech spectrum features extracted from voice files of at least 9 control group (TD: typically developing) speakers and at least 60 speakers with autism spectrum disorder into the support vector machine, and has data classifying the voice into a plurality of autism spectrum disorders. Claim 5 A method for automatically classifying autism spectrum disorder according to claim 1, wherein the pitch characteristics of the voice include the mean, standard deviation, minimum value, maximum value, median, 25th percentile, and 75th percentile of the utterance as sub-characters, the voice quality characteristics of the voice include jitter, shimmer, HNR (Harmonic-to-Noise Ratio), the number of voice breaks, and the ratio of voice breaks as sub-characters, the utterance speed characteristics of the voice include the total length of the utterance, the length of the pause interval within the utterance, the length of the pause-excluded interval within the utterance, speech rate, articulation rate, average syllable length, the number of syllables, and the ratio of the speech sound interval to the total utterance interval as sub-characters, and the voice spectrum characteristics of the voice include log energy and 1st to 12th order MFCC (Mel-frequency cepstral coefficients) values. Claim 6 A computer-readable storage medium for storing a computer program for automatically classifying autism spectrum disorder, wherein the storage medium comprises: a programmed code portion for acquiring and storing meaningful Korean utterances from a test subject and retrieving the stored Korean utterances; a code portion programmed to extract pitch, phonological quality, speech rate, and speech spectrum features from the retrieved Korean utterances using an analysis toolkit; and a code portion programmed to input the extracted pitch, phonological quality, speech rate, and speech spectrum features into a trained machine learning automatic classifier. and includes a code portion programmed to output whether the subject has an autism spectrum disorder from the above-mentioned trained machine learning automatic classifier, wherein the above-mentioned trained machine learning automatic classifier is a Support Vector Machine (SVM) that classifies whether there is an autism spectrum disorder based on the input pitch, phonological quality, speech rate, and speech spectrum features, and wherein the above-mentioned trained machine learning automatic classifier is a Support Vector Machine that classifies speech as autism spectrum disorder and has passed KDSF method training, and the above-mentioned KDSF method comprises: a step of acquiring speech data composed of articulations including all Korean consonant and vowel phonemes from a plurality of speakers including a person with autism spectrum disorder (ASD) by a computer including a processor and memory; a step of preprocessing speech data by a computer including a processor and memory, after removing personal information from the speech data, segmenting the speech into speech units, executing phoneme transcription including spelling transcription, and tagging specific utterances; and a step of, by a computer including a processor and memory, using an analysis toolkit, from the above-mentioned preprocessed speech data, pitch, phonological quality, A step of extracting speech rate and speech spectrum features; a step of inputting the pitch, sound quality, speech rate and speech spectrum features into the support vector machine by a computer including a processor and memory;A computer-readable storage medium comprising: a step of comparing an autism spectrum disorder classification result classified and output by a support vector machine with an autism spectrum disorder classification result input and output after eGeMAPS feature extraction, by a computer including a processor and memory; and a step of acknowledging that it is a trained machine learning automatic classifier when the difference between the two autism spectrum disorder classification results in the upper bound comparison step is within a predetermined error range. Claim 7 delete Claim 8 In claim 6, the analysis toolkit is a computer-readable storage medium comprising librosa for extracting voice spectrum and Parselmouth for extracting pitch, voice quality, and speech rate. Claim 9 A computer-readable storage medium according to claim 6, wherein the trained machine learning automatic classifier inputs pitch, timbre, speech rate, and speech spectrum features extracted from voice files of at least 9 control (TD: typically developing) speakers and at least 60 speakers with autism spectrum disorder into the support vector machine to classify the voice into a plurality of autism spectrum disorders. Claim 10 A computer-readable storage medium according to claim 6, wherein the pitch characteristics of the voice include, as sub-features, the mean, standard deviation, minimum value, maximum value, median, 25th percentile, and 75th percentile of the utterance; the voice quality characteristics of the voice include, as sub-features, jitter, shimmer, HNR (Harmonic-to-Noise Ratio), the number of voice breaks, and the ratio of voice breaks; the speech velocity characteristics of the voice include, as sub-features, the total length of the utterance, the length of the pause interval within the utterance, the length of the pause-excluded interval within the utterance, speech rate, articulation rate, average syllable length, the number of syllables, and the ratio of the speech sound interval to the total utterance interval; and the voice spectrum characteristics of the voice include log energy and 1st to 12th order MFCC (Mel-frequency cepstral coefficients) values.

Images