In order to perform speech recognition well, a huge amount of transcribed speech and textual data in the target language must be available for system training. The high demand for language resources constrains the development ...