One major aspect of current research in speech processing, such as recognition and coding, is to develop systems which retain good performance across a wide variety of acoustic environments. The need for such systems is well appreciated. One method of making robust speech systems is to include a preprocessing stage of noise reduction. The challenge here is to reduce the noise level, preserve or improve the intelligibility and, at the same time, introduce as little distortion as possible. This is a very difficult task and a good solution is still to be found. The work presented here was inspired by previous research by the author in maximum noise reduction of noisy images using Karhunen-Loeve transform.
展开▼
