Robust voiced/unvoiced speech classification with self-organizing maps

The goal of this paper is to show the applicability of a new feature set in voiced/unvoiced (V/UV) classification of speech. The decision is based on the Kohonen-type Self-Organizing Maps (SOM) using this new feature set. The set of input features are computed according to the human auditory system using Warped Linear Prediction (WLP) and found to be robust to background noise - thus the classification is reliable for corrupted speech segments, too. Self-Organizing Maps classify noisy patterns with an error rate of less than 2% at 9 dB signal-to-noise ratio.