Speech Endpoint Detection Based on EMD and Spectral Entropy in Noisy Environments

Abstract:

Accurate endpoint detection is crucial for speech recognition accuracy. A novel approach that finds robust features for endpoint detection based on the empirical mode decomposition (EMD) algorithm and spectral entropy in a noisy environment is proposed. With the EMD, the noise signals can be decomposed into different numbers of sub-signals called intrinsic mode functions (IMFs), which is a zero-mean AM-FM component. Then spectral entropy can be used to extract the desired feature for IMF components. In order to show the effectiveness of the proposed method, we present examples showing that the new measure is more effective than traditional measures. The experiments show that the proposed algorithm can suppress different noise types with different SNR, and the algorithm is robust in the real signal tests.

• Periodical: Applied Mechanics and Materials (Volumes 229-231)
• Pages: 1296-1299
• DOI: https://doi.org/10.4028/www.scientific.net/AMM.229-231.1296
• Online since: November 2012
• Authors: Yan Li Liu, De Xiang Zhang, Ming Wei Ji
• Keywords: Empirical Mode Decomposition (EMD), Endpoint Detection, Noisy Environments, Spectral Entropy

References

[1] G. Tanyer, H. Ozer, IEEE Trans. Speech Audio Processing, Vol. 8 (2000), p.478.

[2] Zhang Yong, Chen Bin. Journal of University of Electronic Science and Technology of China, Vol. 36(2007), p.8.

[3] N. E. Huang, Z. Shen and S. R. Long, Proceedings of the Royal Society London A, Vol. 454 (1998), p.903.

[4] I. Djurovic and L.J. Stankovic, IEEE Transaction on Signal Processing, Vol. 49 (2001), p.2985.

[5] Wang Rangding, Chai Peiqi. Information and Control (In Chinese), Vol. 33 (2004), p.77.

[6] De-Xiang Zhang, Xiao-Pei Wu, and Zhao Lv. Journal of Electronic Science and Technology, Vol. 8 (2010), p.183.