Part of Advances in Neural Information Processing Systems 9 (NIPS 1996)
Steve Waterhouse, Gary Cook
This paper investigates a number of ensemble methods for improv(cid:173) ing the performance of phoneme classification for use in a speech recognition system. Two ensemble methods are described; boosting and mixtures of experts, both in isolation and in combination. Re(cid:173) sults are presented on two speech recognition databases: an isolated word database and a large vocabulary continuous speech database. These results show that principled ensemble methods such as boost(cid:173) ing and mixtures provide superior performance to more naive en(cid:173) semble methods such as averaging.