Rule acquisition for a new speech translation method with waveforms using inductive learning

摘要

Conventional speech translation is a integration of the three processing techniques that contain several complicated parts. Generally, realizing the speech translation had a condition that speech had been transformed to the characteristics. However, achieving high translation performance is like combination among all processings. It is very difficult because each processing is so complicated and has many problems. Then, we propose a new speech translation method that independent on any characteristics of all languages. Our approach hiss some advantages that we use time-varying characteristics of speech without transform to characteristics and we would prevent declining the precision of translation by omitting some processings that were indispensable for traditional methods: morphological analysis, syntactic analysis and others.