Abstract | ||
---|---|---|
In this paper, we present a greedy EM (GEM) method for training Gaussian mixture density (GMD) based acoustic models. In the proposed approach, starting from a single Gaussian, GMD is built up by sequentially adding new components. Each new component is globally selected to avoid local optima. The sequential procedure offers more control over the model structure to achieve better coverage of data. GEM also provides a natural way of integrating information criterion for model complexity selection. Experimental results on WSJ task show that the new method performs consistently better than the conventional method in speech recognition word error rate. |
Year | DOI | Venue |
---|---|---|
2005 | 10.1109/ICASSP.2005.1415209 | ICASSP (1) |
Keywords | Field | DocType |
optimisation,acoustic model training,greedy expectation maximisation method,speech recognition,speech recognition word error rate,learning (artificial intelligence),computational complexity,acoustic signal processing,gaussian mixture density,gaussian processes,model complexity,information criteria,error statistics,greedy em method,hidden markov models,training data,word error rate,computer science,statistics,learning artificial intelligence | Mixture distribution,Pattern recognition,Information Criteria,Computer science,Local optimum,Word error rate,Gaussian,Artificial intelligence,Gaussian process,Acoustic model,Computational complexity theory | Conference |
Volume | ISSN | ISBN |
1 | 1520-6149 | 0-7803-8874-7 |
Citations | PageRank | References |
0 | 0.34 | 3 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Rusheng Hu | 1 | 21 | 2.54 |
Xiaolong Li | 2 | 362 | 36.92 |
Yunxin Zhao | 3 | 807 | 121.74 |