Mapping based noise reduction for robust speech recognition
This project aims at proposing a new noise reduction technique for speech recognition purposes. The proposed method called mapping based noise reduction is performed on the feature vectors extracted from speech signals. In this work we exploit the dimensionality reduction functionality of algorithms such as Locally Linear Embedding and Principal Component Analysis to map the corrupted speech feature vectors to their corresponding noise-free feature vectors. The feature vectors are first mapped to the lower dimensional space and in this space the nearest clean vector to each noisy vector is found, mapped back again to the original space and given as the input to the speech recognition system. We have examined our approach on the speech signals with the artificially added wind noise with different signal to noise ratio values and articulated by two different speakers.