Comparison of various feature decorrelation techniques in automatic speech recognition

Abstract

The design of an optimum front-end module for an automatic speech recognition system is still a great effort of many research teams all over the world. Prepared paper wants to contribute partly to these discussions. It is especially aimed at feature decorrelation techniques based on Maximum Linear Likelihood Transform (MLLT) applied at a different level of matrix clustering. Also the comparison of the MLLT with other decorrelation techniques will be discussed.

Keywords

• decorrelation techniques
• Maximum Linear Likelihood Transform
• speech parameterization