Japanese Dictation Toolkit. 1997 version.

Abstract

The Japanese Dictation Toolkit has been designed and developed as a baseline platform for Japanese LVCSR (Large Vocabulary Continuous Speech Recognition). The platform consists of a standard recognition engine, Japanese phone models and Japanese statistical language models. We set up a variety of Japanese phone HMMs from a contextindependent monophone to a triphone model of thousands of states. They are trained with ASJ (The Acoustical Society of Japan) databases. A lexicon and word N-gram (2-gram and 3-gram) models are constructed with a corpus of Mainichi newspaper. The recognition engine JULIUS is developed for evaluation of both acoustic and language models. As an integrated system of these modules, we have implemented a baseline 5,000-word dictation system and evaluated various components. The software repository is available to the public. +1

A Noise-Robust Continuous Speech Recognition System Using Block-Based Dynamic Range Adjustment

“…Our target is noise-robust Japanese character string recognition. Compared with [20] and [21], the both paper are aiming to Japanese digital strings recognition. The training database and language model are different as well.…”

Improvement of Sound Quality on the Body Conducted Speech Using Differential Acceleration

“…The proposed system would be able to estimate the boundary of a sub-word or a word using the recognition decoder. The continuous sub-word unit recognition decoder is made by Julian that is the Large vocabulary continuous speech recognition system (Kawahara et al, 1999;Lee et al, 2001). Our subjects are Japanese, so we choose the Mora unit as a sub-word.…”

“…Table 2 shows the experimental environments for isolated word recognition. The recognition decoder, Julius (Kawahara et al, 1999;Lee et al, 2001), was used in this experiment. Because Julius is a decoder for large vocabulary continuous speech recognition, it can be changed into isolated word recognition.…”

Body-Conducted Speech Recognition and Application to Speech Support System

“…This decoding system, constructed using the Julian/Julius tools, is known as Japanese Large Vocabulary Continuous Speech Recognition (LVCSR) (Kawahara et al, 1999). Although the Julius speech recognition engine needs a language model, our decoding system does not.…”