Interspeech'2005 - Eurospeech
A generalization of the Discrete Wavelet Packet Transform (DWPT), referred to as Overlapping Discrete Wavelet Packet Transform (ODWPT), is proposed. In contrast to the traditional DWPT, the ODWPT assumes overlapping among the frequency sub-bands at various levels of the transform. Based on this overlapping strategy, a new set of speech features that is specially designed for speaker recognition is derived. The practical significance of our approach has been evaluated in comparative experiments performed on the 2001 NIST Speaker Recognition Evaluation database. The proposed feature set has proved to outperform the widely-used Mel-frequency cepstral coefficients (MFCCs) and other wavelet packet-based speech features that were successfully utilized for speaker recognition.
Bibliographic reference. Siafarikas, Mihalis / Ganchev, Todor / Fakotakis, Nikolaos / Kokkinakis, George (2005): "Overlapping wavelet packet features for speaker verification", In INTERSPEECH-2005, 3121-3124.