A robust word boundary detection algorithm for variable noise-level environment in cars

Abstract

This paper discusses the problem of automatic word boundary detection in the presence of variable-level background noise in cars. Commonly used robust word boundary detection algorithms always assume that the background noise level is fixed and sets fixed thresholds to find the boundary of word signal. In fact, the background noise level in cars varies in the procedure of recording due to speed change and moving environment, and some thresholds should be tuned according to the variation of background noise level. This is the major reason that most robust word boundary detection algorithms cannot work well in the condition of variable background noise level. To solve this problem, we propose a minimum mel-scale frequency band (MiMSB) parameter which can estimate the varying background noise level in cars by adaptively choosing one band with minimum energy from the mel-scale frequency bank. With the MiMSB parameter, some preset thresholds used to find the boundary of word signal are no longer fixed in all the recording intervals. These thresholds will be tuned according to the MiMSB parameter. We also propose an enhanced time frequency (ETF) parameter by extending the time-frequency (TF) parameter proposed by Junqua et al. from single band to multiband spectrum analysis, where the frequency bands help to make the distinction between speech signal and noise. The ETF parameter can extract useful frequency information by choosing some bands of the mel-scale frequency bank. Based on the MiMSB and ETF parameters, we finally propose a new robust algorithm for word boundary detection in variable noise-level environment. The new algorithm has been tested over a variety of noise conditions in cars and has been found to perform well not only under variable background noise level condition, but also under fixed background noise level condition. The new robust algorithm using the MiMSB and ETF parameters achieved higher recognition rate than the TF-based robust algorithm, which has been shown to outperform several commonly used algorithms, by about 5% in variable background noise level condition. It also reduced the recognition error rate due to endpoint detection to 25%, compared to an average of 34% obtained with the TF-based robust algorithm.