From Machine Translation to Automatic Spoken Language Interpretation
Dieter Huber · 1994
Research in automatic spoken language interpretation (synonymous names sometimes used for the same subject area include speech or speech-to-speech translation, automatic telephone interpretation and interpreting telephony) aims at the development of new and advanced information processing techniques that will allow speakers of different languages to converse with one another in their own respective mother tongue via a (stationary or mobile) computer system that automatically translates their speech utterances. The constituent technologies required for such a system include automatic speech recognition (ASR) and speaker adaptation (SpA) at the input side, machine translation (MT) for the actual transfer of meaning between the source and the target language(s), and finally speech synthesis (SS) and voice conversion (VC) to generate the appropriate speech output in the target language(s). Albeit commercial systems that perform these various tasks already exist, mere concatenation of these modules (however well they may perform on their own in conventional applications) is not enough to provide for fully functional speech translation. What is required of each of these components as parts of an automatic speech-to-speech