Abstract
In this thesis, we propose two new features for the classification of a speech frame into a voiced or an unvoiced frame. The first one is the zero crossing rate of a frame after rainfall. This helps to classify a silence frame from speech frames. The second feature is the zero crossing rate of the double autocorrelation function of a frame. This helps to classify a voiced frame from speech frames. The zero crossing rate of the double autocorrelation function is also shown more robust than the ordinary zero crossing rates. The zero crossing rates after rainfall is somewhat a replacement of the feature energy for the classification of speech / non-speech frames. These two features are extensive tested by the TIMIT speech database, and the correct rate is 92.4%.