Using bi-directional associative memories for speaker identification

BAMs (bi-directional associative memories) are simple computational schemes for implementing neural networks. Since computations with BAMs are simple and fast, it is likely that BAMs can be adapted for real time applications and that BAMs can be implemented with few lines of code and therefore may be suitable for TSR (transient and stay resident) applications. In this paper, a set of experiments are described, in which voice samples are associated with names using a modified BAM. A certain set of voice samples are taken by 'trainers' and these are associated with their corresponding names. The paper presents an analysis of the results and concludes that PC based systems, while capable of speaker recognition using BAMs, are as yet not fast enough for real time applications.