Comparing different acoustic modeling techniques for mutlilingual boosting

In this paper, we explore how different acoustic modeling tech-niques can benefit from data in languages other than the target language. We propose an algorithm to perform decision tree state clustering for the recently proposed Kullback-Leibler di-vergence based hidden Markov models (KL-HMM) and com-pare it to subspace Gaussian mixture modeling (SGMM). KL-HMM can exploit multilingual information in the form of uni-versal phoneme posterior features and SGMM benefits from a universal background model that can be trained on multilingual data. Taking the Greek SpeechDat(II) data as an example, we show that KL-HMM performs best for small amounts of target language data. Index Terms: Speech recognition, multilingual acoustic mod-eling, under-resourced language