Output details

Automatic Optimization of Speech Decoder Parameters

<22> IEEE SPL is an outlet well suited for new developments with potential commercial exploitation. This paper addresses the problem of tuning the parameters that govern the size of a speech recogniser’s word sequence search and presents the first automatic method for finding these parameters at a given cost and with low computational complexity. These techniques enabled the development of the publicly available juicer decoder (http://juicer.amiproject.org/juicer/) which led to a spin-out company, Koemei (www.koemei.com), funding from ITSLanguage (www.itslanguage.nl) and a project funded by the U-STAR consortium (www.ustar-consortium.com) for building an iPhone-based speech-to-speech translation system (See ASR Impact Case Study).