Int J Performability Eng ›› 2019, Vol. 15 ›› Issue (11): 3081-3089.doi: 10.23940/ijpe.19.11.p28.30813089

Previous Articles     Next Articles

Speech Enhancement Algorithms in Vehicle Environment

Chunli Wanga,*, Yuchen Lia, and Huaiwei Lub   

  1. aSchool of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou, 730070, China;
    bSchool of Mathematics and Physics, Lanzhou Jiaotong University, Lanzhou, 730070, China
  • Submitted on ; Revised on ; Accepted on
  • Contact: * E-mail address: wcl@mail.lzjtu.cn
  • About author:Chunli Wang received her M.S degree in communication and information systems from Lanzhou Jiaotong University in 2008. She is currently a teacher in the Department of Electron and Information Engineering at Lanzhou Jiaotong University. Her research interests include communication and speech signal processing.Yuchen Li received his B.S degree in communication engineering from Lanzhou Jiaotong University in 2018.Huaiwei Lu is currently a professor at Lanzhou Jiaotong University. His research interests include signal processing.

Abstract: In the actual driving process, the driver is in a complex noise interference environment of the vehicle's own mechanical vibration, the passenger dialogue inside the vehicle, and the sound of other equipment. In order to improve driving efficiency and ensure driving safety, the operation of the vehicle equipment is precisely controlled by the voice control system. Aiming at the residual music noise in traditional spectral subtraction, the improved multi-window spectrum estimation algorithm is applied to improve the estimation accuracy of a priori SNR (signal-to-noise ratio). The experimental results show that the algorithm significantly eliminates the music noise. In the case of low SNR, the signal-to-noise ratio gain is improved by 0.64dB. The waveform similarity and speech naturalness are improved after speech enhancement. Furthermore, the current single-microphone voice de-reverberation technology only takes advantage of the information of time domain and frequency domain with the spatial information limitedly utilized, resulting in a difficulty of achieving a better de-reverberation effect. In light of these insufficiencies, we combine the de-reverberation technique with complex cepstrum blind deconvolution, and a simulation experiment is carried out according to the subjective and objective evaluation indexes of the waveform and the effect of de-reverberated voice, proving that the optimized algorithm improves the intelligibility of the de-reverberated voice.

Key words: vehicle environment, speech enhancement, noise, spectral subtraction, multi-window spectrum estimation