| Popis: |
Deaf and hard-of-hearing (DHH) people cannot recognize the sounds of daily life. These sounds are called environmental sounds (ES), and DHH people find it difficult to live without them. Therefore, hearing assistance dogs and products are used to support DHH people. However, no adequate system to recognize ES, which can be handled for anyone, anywhere, and at any time, has been developed. The aim of this study is to improve the quality of life of DHH people by assisting in ES recognition in various situations. In this study, six types of sounds were used for recognition: two types of water flow, a car horn, fridge alarm, engine idling, and none. None is a sound that does not have a specific ES; however, it contains noise such as the natural sounds of daily life. Feature extraction was conducted by focusing on the value of the spectrogram, and each ES type was discriminated using a support vector machine (SVM), Euclidean dis-tance (EUC), and Mahalanobis distance (MH). Among the six ES, SVM had the highest average discrimination rate (83.3%), followed by EUC (41.2%) and MH (43.3%). In addition to the high discrimination rate, SVM was able to discriminate all six types of sounds, which was not possible with the EUC and MH. Refereed/Peer-reviewed Version of record |
