Speaker-Independent Digit Recognition Using a Neural Network with Time-Delayed Connections

Discussed is the capability of a small neural network to perform speaker-independent recognition of spoken digits in connected speech. The network uses time delays to organize rapidly changing outputs of symbol detectors over the time scale of a word, & is data driven & unclocked. New ideas & procedures developed to achieve useful accuracy in a speaker-independent setting are outlined, including improvement of feature detectors, self-recognition of word ends, reduction in network size, & division of speakers into natural classes. Quantitative experiments based on Texas Instruments (TI) digit databases are described.