Method for estimating severity of dysarthria based on deep audio features

Abstract

The invention discloses a method for estimating severity of dysarthria based on deep audio features. The method comprises steps that acoustic features are extracted; the acoustic features are inputtedto the deep neural network with a bottleneck layer, and depth audio features are extracted from the bottleneck layer; the depth audio features are taken as input, and the Baum-Welch algorithm is utilized to train hidden Markov models; and lastly, the depth audio features of test voice samples are sequentially inputted into the four trained hidden Markov models, four output probabilities are obtained through utilizing the Viterbi algorithm, and the category corresponding to the model with the highest output probability is the dysarthria severity, namely, the decision result. The method is advantaged in that the depth audio features are deep transformation features, compared with traditional acoustic features, feature difference of the dysarthria voice can be more effectively described, andthe more excellent effect in estimating the dysarthria severity can be obtained.

Description

technical field [0001] The invention relates to speech processing and deep learning technology, in particular to a method for estimating the severity of dysarthria based on deep audio features. Background technique [0002] Dysarthria is the most common type of speech dysfunction in children. It often manifests as unclear articulation. It can be subdivided into omitted sounds, substitute sounds, distorted sounds and superfluous sounds, all of which affect normal speech communication and often lead to The other party cannot understand what the patient wishes to express. However, because it does not bring obvious appearance defects or physical pain to the patient, this condition is often not detected in time, thereby delaying the best correction time. As children grow older, the time and cost of dysarthria rehabilitation can rapidly increase. Therefore, timely detection of whether children have dysarthria is of great significance for the rehabilitation of children. At prese...

AI Technical Summary

Problems solved by technology

However, due to the shortage of professionals in related fields in our country, a considerable number of patients with dysarthria cannot receive timely treatment

In addition, the assessment method is highly subjective, and different experts are likely to make different assessments of its severity

Existing inspection instruments and methods, such as fiberoptic palatal and throat endoscopy, television fluorescein radiography, laryngeal dynamics, tongue pressure sensor, etc., although the evaluation is more objective and accurate, these instruments are generally expensive and Some diagnostic methods are invasive and cause severe discomfort to the patient, who are very reluctant to cooperate with the diagnosis

Images