Open Access

ARTICLE

Noise Cancellation Based on Voice Activity Detection Using Spectral Variation for Speech Recognition in Smart Home Devices

Jeong-Sik Park¹, Seok-Hoon Kim^2,*

1 Department of English Linguistics & Language Technology, Hankuk University of Foreign Studies, Seoul, Republic of Korea
2 Department of Electronic Commerce, Paichai University Studies, Daejeon, Republic of Korea

* Corresponding Author: Seok-Hoon Kim,

Intelligent Automation & Soft Computing 2020, 26(1), 149-159. https://doi.org/10.31209/2019.100000136

Abstract

Variety types of smart home devices have a main function of a human-machine interaction by speech recognition. Speech recognition system may be vulnerable to rapidly changing noises in home environments. This study proposes an efficient noise cancellation approach to eliminate the noises directly on the devices in real time. Firstly, we propose an advanced voice activity detection (VAD) technique to efficiently detect speech and non-speech regions on the basis of spectral property of speech signals. The VAD is then employed to enhance the conventional spectral subtraction method by steadily estimating noise signals in non-speech regions. On several experiments, our approach achieved superior performance compared to the conventional noise reduction approaches.

---

Keywords

---