Crosslingual and bilingual speech recognition with Slovak and Czech speechdat-e databases

This paper presents the work on crosslingual and bilingual speech recognition carried out with SpeechDat databases for Czech and Slovak language. The work follows the MASPER initiative that was formed as a part of the COST 278 Action. In crosslingual experiments the expert-driven and the datadriven approaches were used for transferring monolingual source acoustic models to a target language. The results’ analysis showed that the crosslingual Czech/Slovak speech recognition performance outperforms the results got in MASPER initiative for other target languages and that similarities between source and target language have a significant influence on the performance. Consecutively a bilingual Czech/Slovak recognition experiment with linked SpeechDat-CZ/SK was performed. The positive results indicate possibility to share Czech and Slovak speech databases for training bilingual and monolingual acoustic models.