Computerized Based Analysis for Detection and Severity Assessment of Stuttering

Overview

In the light of previous attempts to design and develop automated and objective measures for automatic speech recognition system that detects disfluent speech and assess its severity, yet fully automated measurement of stuttered speech is not available. This study was triggered by the need to design and develop a simple and reliable computerized tool for identification of stuttering and measurement for its severity. Therefore, the aim of this study is to develop a user interface that can work on windows system for the adopted stuttering recognition model which can be used in clinical practice by physicians and therapists.

Stuttering is a speech disorder in which the normal flow of speech is disrupted by occurrences of dysfluencies, such as repetition, prolongations and blocks (1). Features that have been found to differ between stutterers and nonstutterers are rate of speech and frequency of dysfluent utterances (2). An Arabic version of stuttering severity instrument (A-SSI) is used to assess the stuttering severity In it, the overall severity score of stuttering is measured by combining the scores of percentages of Stuttered Syllables (%SS), Mean Duration of the Three Longest Stuttering Events (MDTLSE), and Physical Concomitants (PC) (3). The subjective assessment methods of stuttering are; time-consuming, prone to error, subjective (4), so it is better to automate the measurement of disfluencies using speech recognition technologies and computational intelligence (5). Speech recognition executes a task similar to what the human brain undertakes (6). Stuttering detection system has three main steps which are acoustic processing, feature extraction and classification/recognition (7). the speech signals are pre-processed (8), and certain features are extracted from them by signal processing techniques, e.g. Mell frequency cepstral coefficients (MFCC) (9). (MFCC) is considered the most popular used feature extraction technique (10). The classification process contains two steps; training and testing (11). In training process, data is labeled based on the classes and a model is learned. In testing phase: the model is tested and computed the accuracy, sensitivity, and specificity of the classification models (11). Finally, stuttering from non-stuttering speech will be recognized and separated (5) also to assess the severity of stuttered speech.

Conditions

Interventions

Eligibility

Outcomes