In this paper, a speech enhancement method is presented, which uses spectral and time domain processing and achieves a trade-off between effective noise reduction and low computational load for real-time operations. First, a spectral subtraction method is used to reduce the effect of additive broadband noise on speech. Then, a novel weighting function is used to reduce a residual "musical noise" in time domain. This weighting function is a compound of a short-time zero crossing value and a short-time energy of speech signal. For evaluation of improvement of speech recognition the Slovenian SpeechDat FDB, the German SpeechDat FDB and SpeechDat-Car, as well as the Spanish SpeechDat FDB databases together with the HTK recognition toolkit were used. Word recognition accuracy in connected digits recognition task was improved by 8.7% for Slovenian FDB, by 5.1% for Spanish FDB, by 3.2% for German SpeechDat-Car, and by 2% for German SpeechDat FDB database.
Cite as: Kotnik, B., Kacic, Z., Horvat, B. (2001) A computational efficient real time noise robust speech recognition based on improved spectral subtraction method. Proc. 7th European Conference on Speech Communication and Technology (Eurospeech 2001), 1123-1126, doi: 10.21437/Eurospeech.2001-282
@inproceedings{kotnik01b_eurospeech,
author={Bojan Kotnik and Zdravko Kacic and Bogomir Horvat},
title={{A computational efficient real time noise robust speech recognition based on improved spectral subtraction method}},
year=2001,
booktitle={Proc. 7th European Conference on Speech Communication and Technology (Eurospeech 2001)},
pages={1123--1126},
doi={10.21437/Eurospeech.2001-282}
}