Phoneme recognition with a time-delay neural network
A. Shaw, Rebecca Mitchell · 1990
A time-delay neural network architecture is used for speaker-dependent recognition of the long vowel sounds a, e, and i. The following areas are investigated: increasing the amount of data supplied to the network, allowing longer and variable-length utterances, and using English rather than Japanese speech. With this architecture, 100% recognition has been obtained for a population of six speakers