Hidden Markov Models have many applications in signal processing and pattern recognition, but their convergence-based training algorithms are known to suffer from over-sensitivity to the initial random model choice. This paper describes the boundary between regions in which ensemble learning is superior to Rabiner’s multiple-sequence Baum-Welch training method, and proposes techniques for determining the best method in any arbitrary situation. It also studies the suitability of the training methods using the condition number, a recently proposed diagnostic tool for test-ing the quality of the model. A new method for training Hidden Markov Models Correspondence to:

In this paper we present a new adaptation technique for our hybrid large vocabulary continuous speech recognition system. In most adaptation approaches the HMM parameters are reestimated. In our approach, however, we train a speaker independent continuous speech recognizer, then we keep the HMM parameters fixed and we train a second network, which transforms the features of the adaptation data to fit the HMM parameters. Thus, less parameters have to be estimated, and therefore this approach performs well even for a small number of adaptation data. With this approach we achieve relative improvements in recognition rates on the Wall Street Journal (WSJ) task of 16.5%. 1. INTRODUCTION Over the last years we developed a high performance speech recognition system based on a new hybrid approach [5][6]. Recently [7] we could show that the performance of our hybrid connectionist/HMM speaker independent continuous speech recognition systems is very close to the performance of stand...