Modified two-dimensional root cepstrum analysis
Hosein Marvi, E. Chilton · Electronics Letters · 2005
A novel feature vector for speech recognition applications is proposed. These new features are obtained by generalising the two-dimensional root cepstrum (TDRC) using the mel-scaled root filter bank energies of a speech frame in the process of computing the TDRC. Results indicate superior performance over conventional techniques in an isolated word recognition problem.