5th International Conference on Spoken Language Processing
In the context of command-and-control applications, we exploit confidence measures in order to classify utterances into two categories: utterances within the vocabulary which are recognized correctly, and other (out-of-vocabulary= OOV and misrecognized) utterances. We investigate the classification error rate (CER) of several classes of confidence measures and transformations based on a database containing 3345 utterances by 50 male and female individuals, employing data-independent and data-dependent measures. The transformations we investigated include mapping to single confidence measures, LDA-transformed measures, and other linear combinations of these measures. These combinations are computed by means of neural networks trained with Bayes-optimal, and with Gardner-Derrida-optimal criteria. Compared to a recognition system without confidence measures, the selection of (various combinations of) confidence measures, and the selection of suitable neural network architectures and training methods, continuously improves the CER from 16.7% to 6.6% (-60% relative). Furthermore, a linear perceptron generalizes better than a non-linear backpropagation network.
Bibliographic reference. Dolfing, J. G. A. / Wendemuth, Andreas (1998): "Combination of confidence measures in isolated word recognition", In ICSLP-1998, paper 0481.