Speaker Identification with Little Training Data Based on Semiconductor Neurocomputer
Dewen Zhuang, Wenming Cao, Shoujue Wang · 2006
In this article we propose a novel speaker identification method using semiconductor neurocomputer. We also give the multi-weight neuron training and recognition algorithm based on high dimensional feature space vertex covering theory. The average distance of the frame between the test sentence and the speaker covering region are used as our decision rule, then in the condition of little training data we can identify all the speakers in the given test sets.