A denosing method of frequency spectrum for recognition of dashboard sounds
Pengfei Li, Qingxiang Wu, Caiyun Wu, Caiyou Yuan · 2017
In this article a frequency spectrum-based approach is proposed to extract features and identify warning tones from dashboard. The purpose of this paper is to identify the warning tones emitted from the dashboard, such as turn signal lights, unmanned seat belts, position lights, etc. After analysis of the spectral characteristics of sound signals, the information is extracted directly from the spectral characteristics as the basis of classification and identification. Firstly, the standard voices of different instruments are collected as templates. The test sound is compared with the template sounds when testing, and similarity is calculated. When the similarity is higher than a threshold, test sound and template sound are regarded as the same voice. Otherwise that is not qualified sound and the meter is considered to be defective. The usual way to remove noise was in the time domain. A new approach has been proposed to denoise in the spectral space and obtain the efficiently spectral characteristics of the standard sounds. Finally this method is applied to dashboard voice recognition and has achieved good results.