A new spectrum envelope estimation technique based on sample projections

Spectral envelope plays a very important role in different areas of speech applications. This envelope contains the intelligible information within a given message. A new spectral envelope estimation method with two different versions has been introduced in this paper. This method is based on sample projections. The proposed technique incorporates the weighted effect of extrapolation of previous samples on the current sample. This is achieved through the implementation of an Auto Regressive (AR) model. Results showed that higher peaks in the envelope are obtained based on the proposed technique. This means that more emphasis is given on the poles of the filter transfer function and hence leading to better conveying of the message effectively. The proposed technique leads to better speech quality when applied to speech coding and speech recognition. It is found that the filter order resulting from the proposed method will increase by one as compared to the traditional linear prediction (LP) method. Moreover, in some cases, the resulting filter must be stabilized by moving the poles to lie within the unit circle.