EUROSPEECH 2001 Scandinavia
7th European Conference on Speech Communication and Technology

Aalborg, Denmark
September 3-7, 2001


Objective Evaluation of Methods for Quantization of Variable-Dimension Spectral Vectors in WI Speech Coding

Jani Nurminen (1), Ari Heikkinen (2), Jukka Saarinen (1)

(1) Tampere University of Technology, Finland
(2) Nokia Research Center, Finland

In this paper, we present a comprehensive evaluation of five quantization techniques for variable-dimension spectral vectors in a waveform interpolation speech coder. Each technique included in the evaluation is based on dimension conversions. The conversions are performed using zero-pad and truncation, frequency bins, band-limited interpolation, discrete cosine transform, and polynomial approximation. In addition to assessing quantization accuracy, this study considers the complexity of the techniques. The evaluation indicates that the selection of the optimal quantization technique is a trade-off between coding accuracy, complexity, and memory requirements. According to our results, the technique based on discrete cosine transform appears to be a strong candidate for many applications.

Full Paper

Bibliographic reference.  Nurminen, Jani / Heikkinen, Ari / Saarinen, Jukka (2001): "Objective evaluation of methods for quantization of variable-dimension spectral vectors in WI speech coding", In EUROSPEECH-2001, 1969-1972.