AI-based Speech Analysis Method and System for Diagnosing Language Development Delay in Korean Children

Abstract

The present invention relates to an AI-based speech analysis method and system for the early diagnosis of language development delays in Korean children. The system collects children's speech using segmentation-indicating and long-term recording methods, and evaluates pronunciation, sentence structure, and vocabulary diversity by optimizing the latest speech recognition models, such as Whisper and Wav2Vec2.0, for the speech patterns of Korean children. It provides a system that determines the presence of language development delays by calculating the evaluation results into quantified scores and supports in-depth examinations by experts. Through this, effects such as reduced testing time, improved accuracy, enhanced early detection rates, and a reduction in unnecessary in-depth examinations can be achieved.

Description

Technology Field

[0001] The present invention relates to the field of language development diagnostic technology utilizing artificial intelligence-based voice analysis technology, and in particular to a technology that analyzes a child's voice data to diagnose language development delays early and support in-depth testing. Background Technology

[0002] Delayed language development in children has serious consequences, with 50–70% experiencing learning disabilities or reading and writing disorders during the school years, and about 40% struggling to form peer relationships. According to the American Speech-Language and Hearing Association (ASHA, 2020), early intervention before the age of five is most effective.

[0003] The checklist-based evaluation methods currently in use have the following problems:

[0005] 1. Evaluation distortion caused by subjective checklists completed by parents or teachers

[0006] 2. It takes an average of 30 minutes just to complete the checklist, and additional time is required for results analysis and expert referrals.

[0007] 3. There are many cases where children identified as having 'language delay' actually have no problems, and conversely, there are frequent instances where actual language development delays are confirmed among children evaluated as 'normal'.

[0009] Existing AI-based voice analysis technologies have limitations in that they cannot accurately reflect the unique language development characteristics of children because they were primarily developed to suit the voice characteristics of adults. The problem to be solved

[0010] The present invention aims to overcome the limitations of existing checklist methods and adult-centered AI voice analysis models and to solve the following problems:

[0012] 1. Development of an AI voice analysis model specialized for Korean children

[0013] 2. Providing technology capable of comprehensively evaluating pronunciation, sentence structure, vocabulary diversity, etc.

[0014] 3. Reduced testing time (from 30 minutes to within 10 minutes) while improving accuracy (up to 35%)

[0015] 4. Provide tools to support in-depth expert testing based on voice analysis results means of solving the problem

[0016] The AI-based voice analysis method of the present invention comprises the following steps:

[0018] 1. Child voice collection in two ways: segmented instruction type and long-term recording type:

[0019] - Classification Indicative Type: Evaluates pronunciation, sentence structure, and vocabulary diversity through responses to specific questions

[0020] - Long-term recording type: Analyzes the child's natural language use to determine developmental level

[0022] 2. Optimizing the latest speech recognition models, such as Whisper and Wav2Vec2.0, for Korean children's speech patterns:

[0023] - MFCC (Mel-frequency cepstral coefficients) based feature extraction

[0024] - Application of a hybrid deep learning model combining CNN, LSTM, etc.

[0025] - Development of an algorithm specialized in analyzing children's pronunciation errors, phonological patterns, and grammatical errors

[0027] 3. Perform multidimensional language analysis:

[0028] - Acoustic analysis: Analysis of formants, pitch, speech velocity, etc.

[0029] - Pronunciation Accuracy: Evaluation of syllable-level speech accuracy

[0030] - Sentence Structure: Analysis of grammatical complexity and word order accuracy

[0031] - Vocabulary diversity: Evaluation of the diversity and appropriateness of vocabulary used

[0033] 4. Implementation of a 4-Step Automated Diagnostic Process:

[0034] - Step 1: Child Voice Collection

[0035] - Step 2: AI Automatic Analysis

[0036] - Step 3: Determine if there is a language development delay

[0037] - Step 4: Support for in-depth expert testing if necessary

[0039] 5. Advanced Testing Support System for Professionals:

[0040] - Data visualization of AI analysis results

[0041] - Provides acoustic analysis data

[0042] - Automated report generation based on objective data Effects of the invention

[0044] The following effects can be obtained through the present invention:

[0046] 1. 50% reduction in inspection time compared to the existing checklist method (30 minutes)

[0047] 2. Up to 35% improvement in screening accuracy through objective data-based analysis

[0048] 3. Overcoming the limitations of subjective evaluation to improve the early detection rate of language development problems by more than 20%

[0049] 4. Cost savings achieved by reducing unnecessary in-depth testing by 30% through accurate screening.

[0050] 5. Maximizing diagnostic efficiency through an AI and expert collaboration system Brief explanation of the drawing

[0052] Figure 1 is an overall flowchart of the AI-based speech analysis method for diagnosing language development delay in Korean children according to the present invention. Figure 2 is a schematic diagram illustrating the four-step diagnostic process of the present invention. Figure 3 is a structural diagram of the advanced inspection support system for experts according to the present invention. Specific details for implementing the invention

[0053] [Example 1: Development of an AI Model Specialized for Korean Children]

[0054] The present invention collects speech data of Korean children aged 3 to 5 in cooperation with Durubareun Social Cooperative and develops a speech recognition model specialized for Korean children based on this data. Features are extracted from the collected speech data using the MFCC technique, and fine-tuning is performed based on the Whisper model and the Wav2Vec2.0 model. In particular, a data augmentation technique that considers phonological characteristics is applied to reflect the unique pronunciation error patterns, sentence structure characteristics, and age-specific vocabulary usage patterns of Korean children into the model.

[0056] [Example 2: Hybrid Inspection Approach]

[0057] This invention implements a hybrid approach that combines a basic screening test (5-10 minutes) with optional extended recording (1-2 hours). The basic screening test evaluates a child's fundamental language abilities through structured tasks, and if in-depth analysis is required, additional recordings are taken in a home or educational setting to analyze natural language usage patterns. This hybrid approach enables the simultaneous assurance of both practicality and precision in the test.

[0059] [Example 3: Advanced Examination Support Tool for Experts]

[0060] The professional in-depth examination support tool of the present invention provides visualized AI analysis results. It presents acoustic analysis (formants, pitch, speech rate, etc.) and language pattern analysis data in the form of an intuitively understandable dashboard, and provides objective data and automatically generated analysis reports that allow for a quick overview of a child's language development status. Through this, professionals can perform more efficient and accurate diagnoses during the in-depth examination process.

[0062] [Example 4: Implementation of Cloud-based Web and Mobile Platform]

[0063] The present invention is implemented as a cloud-based web and mobile platform accessible anytime and anywhere. Users can record a child's voice and check the AI ​​analysis results in real time via a web browser or mobile application. Furthermore, it is designed to enable integration with external systems via APIs, allowing it to be used in conjunction with existing systems in hospitals, speech rehabilitation centers, educational institutions, and the like. Explanation of the symbols

[0065] 100: Voice data input section 110: Child-Specialized AI Analysis Department 120: Phonological Error Detection Unit 130: Speech Rate Analysis Unit 140: Sentence Structure Evaluation Section 150: Vocabulary Diversity Evaluation Department 160: Scoring and Automatic Report Generation Section 170: Expert Advanced Testing Support Department 180: API Integration Module

Claims

Claim 1 A method for diagnosing language development delay in Korean children based on an AI-based speech analysis method, comprising: a step of collecting the child's speech using a segmentation-reference method and a long-term recording method; a step of performing feature extraction and data preprocessing using Mel-frequency cepstral coefficients (MFCC) on the collected speech data; a step of constructing an AI model optimized for the speech patterns of Korean children using the latest speech recognition models such as Whisper and Wav2Vec2.0; a step of evaluating pronunciation, sentence structure, and vocabulary diversity through the AI ​​model; a step of determining whether there is a language development delay by calculating the evaluation results into a quantified score; and a step of generating an automated screening test report based on the results. Claim 2 A method according to claim 1, characterized in that the above-mentioned distinction-indicating method evaluates pronunciation, sentence structure, and vocabulary diversity through responses to specific questions, and the above-mentioned long-term recording method analyzes the developmental level by recording the child's natural language use for 1-2 hours. Claim 3 A method according to claim 1, characterized in that the AI ​​model analyzes specific phonological error patterns, pronunciation accuracy, grammatical complexity, and vocabulary diversity of Korean children using a hybrid deep learning model combining a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM). Claim 4 A method according to claim 1, wherein the AI ​​model distinguishes detailed stages of language development through formant analysis, pitch change, speech velocity analysis, etc. by applying acoustic feature extraction technology. Claim 5 A method according to claim 1, further comprising a step of supporting an expert's in-depth examination based on the evaluation results, wherein the support for the in-depth examination improves the expert's diagnostic accuracy and efficiency through the visualization of AI analysis result data, the provision of acoustic analysis data, and the generation of an automatic report based on objective data. Claim 6 A method according to claim 1, further comprising the step of providing services to speech rehabilitation centers / hospitals (B2B), education offices / public health centers (B2G), and parents (B2C) through a cloud-based web / mobile platform; and the step of providing an API integration module for integration with external systems. Claim 7 A Korean language child language development delay diagnosis system comprising: a voice data input unit receiving a child's voice; a child-specific AI analysis unit including an AI model optimized for the speech patterns of Korean children; an evaluation unit evaluating pronunciation, sentence structure, and vocabulary diversity; a scoring and report generation unit generating an automated report by calculating the evaluation results into a quantified score; and an advanced test support unit supporting advanced testing by experts.

Images