Introduction to the special issue on processing under-resourced languages

Abstract

The creation of language and acoustic resources, for any given spoken language, is typically a costly task. For example, a large amount of time and money is required to properly create annotated speech corpora for automatic speech recognition (ASR), domain-specific text corpora for language modeling (LM), etc. The development of speech technologies (ASR, Text-to-Speech) for the already high-resourced languages (such as English, French or Mandarin, for example) is less constrained by this issue and, consequently, high-performance commercial systems are already on the market. On the other hand, for under-resourced languages, the above issue is typically the main obstacle.