Adaptive Non-Linear Prediction for Speech Signals in Mixture Noise Environments

For robust prediction analysis for speech signals in impulsive noise environments, we had proposed the OSLMS with AS predictor in Y. Ohhashi et al. (2004). Assuming general noise environments, however, we cannot neglect background noises. Therefore, in this paper, we saddled properties of the LPC in a mixture noise environment which has a content of impulsive noises and white Gaussian noises, and prove that an adaptive predictor perform more effectively in the mixture noises which has white noises with low power than batch predictors. Furthermore, we conducted prediction experiments using 3 adaptive predictors including the OSLMS with AS in the mixture noise environments. As a result, She OSLMS with AS predictor provided good performance in mixture noise environments which has practical white noise power, too