サフィックス・アレイに基づく言語モデルを用いた音声認識に関する研究

For obtaining high speech recognition performance, we need high quality acoustic model and language
model of speech recognition. In this study, we focus on the language model. The conventional language
models, which are CFG, N-gram model, and so on, have some problems which are outputted the
non-language characters and words sequence. Therefore, in this paper, we proposed the language
model which was used the suffix array for speech recognition. The suffix array was proposed for the
information retrieval. The advantages of the suffix array were that “予測可能” “無駄な仮説が生成さ
れない” For evaluating the proposed language model, we conducted the similarly music information
retrieval experiment using MIDI database. The experimental results showed that the proposed method
was useful for the music information retrieval.