The Korean Large Vocabulary Continuous Speech Recognition Platform

For educational and research purposes, we design and evaluate a Korean speech recognitionplatform to build a decoder. The platform has an object‐oriented architecture so that researcherscan modify the platform easily and evaluate the performance of a recognition algorithm of theirinterests. The platform has the following functionalities: Noise reduction, speech detection,feature extraction, hidden Markov model (HMM)‐based acoustic modeling, cross‐word modeling, ngram language modeling, n‐best search, word graph generation, and Korean‐specific languageprocessing. The platform can handle both lexical search trees for large vocabulary speechrecognition, and finite‐state networks for small‐tomedium vocabulary speech recognition. Itperforms the word‐dependent n‐best search algorithm with a bigram language model in thefirst forward search stage, then extracts a word lattice, and finally rescores the lattice with atrigram language model in the second backward search stage. In a large vocabulary continuousspeech recognition task, we compare the performance of the platform with HTK and Julius.