Developments of Machine Learning Schemes for Dynamic Time-Wrapping-Based Speech Recognition

Abstract

This paper presents a machine learning scheme for dynamic time-wrapping-based (DTW) speech recognition. Two categories of learning strategies, supervised and unsupervised, were developed for DTW. Two supervised learning methods, incremental learning and priority-rejection learning, were proposed in this study. The incremental learning method is conceptually simple but still suffers from a large database of keywords for matching the testing template. The priority-rejection learning method can effectively reduce the matching time with a slight decrease in recognition accuracy. Regarding the unsupervised learning category, an automatic learning approach, called "most-matching learning, " which is based on priority-rejection learning, was developed in this study. Most-matching learning can be used to intelligently choose the appropriate utterances for system learning. The effectiveness and efficiency of all three proposed machine-learning approaches for DTW were demonstrated using keyword speech recognition experiments.

An eigenspace-based method with a user adaptation scheme for human gesture recognition by using Kinect 3D data

“…The developed approach, called Eigen3Dgesture, involves using PCA to derive significant eigen3Dgestures possessing the most critical information from Kinect 3D data. In addition, to enhance the proposed approach and the performance of the constructed eigenspace of Kinect 3D data, a user adaptation (UA) scheme [17], [18], which entails employing the active gesture data of a test user to adjust the eigenspace of Kinect 3D data such that the Eigen3Dgesture recognition model is more representative of a new test user, was developed in this study. Studies regarding the use of adaptation schemes in HMM-based and DTW-based Kinect gesture recognition systems are extremely rare.…”

An HMM-Like Dynamic Time Warping Scheme for Automatic Speech Recognition

“…The testing speaker utters the same words for the operated voice command, but these uttered commands will not have exactly the same result so that speech recognition with the correct recognition outcome in each recognition test will be hard to achieve. To overcome this problem, related works on speech recognition enhancements have been quite common in the recent years, and most of those studies aimed at increasing the reliability of the recognition result by improving the recognition system [4] or reducing the mismatch phenomenon between a new speaker and the speech recognition system by performing machine learning schemes [5] or adaptive designs [6] on original speech recognition system.…”

“…In the author's previous work [5], speaker learning for DTW speech recognition has been explored where the learning strategy is interpolated into traditional DTW. Under the scheme, the DTW system is additionally equipped with the developed machine learning approaches for modifying the database containing referenced templates of speech patterns [5].…”

“…In the author's previous work [5], speaker learning for DTW speech recognition has been explored where the learning strategy is interpolated into traditional DTW. Under the scheme, the DTW system is additionally equipped with the developed machine learning approaches for modifying the database containing referenced templates of speech patterns [5]. However, the fundamental structure of DTW in [5] is almost still the same as that of conventional DTW except the additionally given machine learning scheme for the database of DTW referenced templates, both of which still belong to feature-based recognition techniques.…”

“…Under the scheme, the DTW system is additionally equipped with the developed machine learning approaches for modifying the database containing referenced templates of speech patterns [5]. However, the fundamental structure of DTW in [5] is almost still the same as that of conventional DTW except the additionally given machine learning scheme for the database of DTW referenced templates, both of which still belong to feature-based recognition techniques. The DTW system learning performance by the developed work in [5] will still be largely restricted due to the essence of invariable feature-based template matching and the lack of a statistical recognition model when performing recognition.…”