Clustering of triphones using phoneme similarity estimation for the definition of a multilingual set of triphones

摘要: This paper addresses the problem of multilingual acoustic modelling for design speech recognisers. An agglomerative clustering algorithm definition set triphones is proposed. based on an indirect distance measure defined as a weighted sum explicit estimates context similarity monophone level. The estimation method Houtgast. new was tested in recognition experiment three languages. applied monolingual triphone sets language specific recognisers all In order to evaluate algorithm, performance compared reference system composed operating parallel, and produced by tree-based algorithm. All experiments were 1000 FDB SpeechDat(II) databases (Slovenian, Spanish German). Experiments have shown that use results significant reduction number with minor degradation rate.