Speech recognition tool

Abstract

This project is to study the use of various mathematical or statistical methods and techniques used in speech recognition to identify defects in an aircraft material. Aircraft body is likely to have defects after long hours of flights and it is of utmost importance to detect any defects. There are 24 defects and 1 non defect data inputs in this study and they are presented in a form of sound waves. Therefore techniques used in speech recognition are being explored here. Some of the techniques to be explored are such as Earth Mover’s Distance, Correlation and First Derivative Correlation search, Linear Discriminant Analysis, Dynamic Time Warping and Neural Network. Earth Mover’s Distance can be seen employed in colour and image recognition. It is a kind of dissimilarity assessment technique which is applicable to multi-dimensional distributions. Correlation and First Derivative Correlation Search is a type of statistical measure for relationships between distributions and the result of it will show how related the data is from each other. Linear Discriminant Analysis is often employed in statistics and machine learning. It is most often used to find separation between classes of objects. Dynamic Time Warping is a technique used to measure similarity between two distributions and this method is able to analyse data and compare them disregarding the time and speed difference. Neural Network is a complex analytical method which creates a network from training it. With it, Neural Network is able to recognise trends and patterns from data which humans may be unable to. With these few methods, the aim is to study the plausibility and effectiveness of applying these techniques into defect detection for aircraft material.