Learning spectral-temporal dependencies using connectionist networks

Describes the application of a layered connectionist network for continuous digit recognition using syllable based segmentation. Knowledge is distributed over many processing units. The behavior of the network in response to a particular input pattern is a collective decision based on the exchange of information among the processing units. A supervised back-propagation learning algorithm is used to repeatedly adjust the weights in the network, to minimize the difference between the actual output vector and the desired output vector. The performance of the network is compared to that of a nearest neighbor classifier trained and tested on the same database. Speaker-dependent continuous digit recognition experiments were performed using a total of 540 digit strings with an average length of 4 digits, collected from six speakers (4 male and 2 female)