Scaly artificial neural networks for speaker-independent recognition of isolated words
Anna Krause, Heidi Hackbarth · International Conference on Acoustics, Speech, and Signal Processing · 2003
Artificial neural networks were investigated as an approach to speaker-independent recognition of isolated words from a variable-size vocabulary. A scaly pattern of connecting links between the input and hidden layer was established within a strict feed-forward wiring structure. The amount of training material and vocabulary size as well as the choice of network dimensions, such as the number of layers and input and hidden units, are discussed. Experimental results are reported and compared to the recognition performance obtained with a fully connected layered network. Error rates lower than 5% for vocabularies with up to 20 words suggest that appropriately designed artificial neural networks are well suited for a speaker-independent recognition task.>