| Summary: | 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.
|
|---|