Bloomfield Model Based Signal Process for Networks

This paper proposes a novel speech signal analysis approach based on the Bloomfield ( $BF$ ) model, and provides a formulation of a time-domain $BF$ model for speech signals with which speech signals can be reconstructed and the relevant characteristic parameters analyzed. The relationship between the parameters of the $BF$ model and those of the linear prediction ( $LP$ ) model are derived, and the speech feature sets derived via the $LP$ and $BF$ models are compared. A new algorithm is proposed for the recognition of isolated digit speech that utilizes a vector quantization approach and is based on the $BF$ Model. The result is obtained with this $BF$ approach that provides better results than those of the $LP$ model when predicting speech signals. In particular, the $BF$ approach has several advantages, including fewer parameters, a lower computational complexity, and accurate characterization of speakers. These advantages ensure the utility of the $BF$ model in speech processing applications.