Voice Reading Assessment Using MFCCs for Impaired Speech
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for assessing cognitive and speech motor impairments, particularly in conditions like Huntington's Disease, are time-consuming and require clinical expertise, and existing speech analysis technologies lack sensitivity and accuracy, especially in tasks like the Stroop test, and are not language-independent.
Innovation Solution
A model-free, language-independent method for analyzing voice recordings from word-reading tests using Mel-frequency cepstral coefficients (MFCCs) and power Mel-spectrograms to segment words or syllables, enabling accurate counting of correctly read words and determining speech rate, suitable for self-assessment by individuals with speech impairments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If speech analysis software using oscillographic sound pressure signal or automated syllable detection is used, then speech rate can be measured, but sensitivity is insufficient for fast speech in word-reading tasks and correctness cannot be assessed
Solution Approach 1:
The patent replaces traditional mechanical/acoustic analysis methods (oscillographic sound pressure signal analysis) with speech recognition technology using deep learning models. This substitution enables accurate identification of individual words in fast speech sequences, allowing both speech rate measurement and correctness assessment that were not achievable with conventional acoustic methods.
Solution Approach 2:
The patent transforms the approach by changing from analyzing continuous acoustic signals to processing discrete speech recognition results. By converting the problem from acoustic parameter analysis to linguistic unit identification, the system can accurately measure speech rate in words per second and assess correctness for fast speech tasks.
2Ease of operation
If pre-trained speech recognition models built on healthy populations are used, then speech content can be understood, but accuracy is poor for patients with speech impairments
Solution Approach 1:
The patent adapts speech recognition parameters and model characteristics to accommodate impaired speech patterns. By adjusting the recognition system to handle variations in speech production caused by neurological conditions, the system maintains high accuracy for both healthy and impaired speakers without requiring separate models for each condition.
Solution Approach 2:
The patent creates a universal speech recognition system that functions accurately across diverse populations including healthy individuals and patients with various speech impairments. The system is designed to handle multiple speech patterns and conditions within a single framework, eliminating the need for population-specific models.
3Adaptability or versatility
If automated speech recognition is implemented, then language independence can be achieved, but computational complexity and processing requirements increase
Solution Approach 1:
The patent replaces traditional language-specific acoustic modeling with deep learning-based speech recognition that automatically adapts to different languages. This substitution eliminates the need for separate language models and manual linguistic processing, achieving language independence while reducing overall system complexity despite the computational intensity of deep learning.
Data Source
AI summary
The application relates to assessing cognitive impairment and/or speech motor impairment. The method comprises analysing a voice recording from a word-reading test by identifying a plurality of segments of the voice recording that correspond to single words or syllables and determining the number of correctly read words in the voice recording and/or the speech rate. Determining the correct number of words in the recording may comprise computing one or more Mel-frequency cepstral coefficients for the segments, clustering the resulting vectors of values into n clusters, wherein each cluster has n possible labels, predicting a sequence of words in the voice recording using the labels associated with the clustered vectors of values, performing a sequence alignment between the predicted sequence of words and the sequence of words used in the word reading test, selecting the labels that result in the best alignment and counting the number of matches in the alignment.