Computational Reduction of Contenous Speech Recognition Software "Julius" on SuperH Microprocessor

PC向け連続音声認識プログラムJuliusのSuperHマイコン(SH-4A)への搭載に関して行った処理の高速化と,評価実験について報告する.計算リソースの限られたマイコン上で動作させるため,仮説探索時のメモリ管理の最適化や音響尤度計算の高速化を実施した.語彙数5,000語での評価実験では,最適化前のJuliusの実行速度に対して3.7倍の高速化を実現し,SH-4A上での実時間動作を達成した.また,語彙数20,000単語での評価でも実時間の1.25倍で動作すること確認した.最後に,応用アプリケーションとしてT-Engine上に実装した質問応答システムについて報告する.To expand CSR (continuous speech recognition) software to the mobile environmental use, we have developed embedded version of Julius (Embedded Julius). In this paper, we describe an implementation of the "Embedded Julius" on a SH-4A microprocessor. SH-4A is a high-end MPU with on-chip FPU. However, further computational reduction is necessary for the CSR software to operate real-time. Applying some optimizations (efficient memory management, modified GMS), the "Embedded Julius" achieves real-time processing on the SH-4A. The experimental results show 0.73 x real-time, resulting 3.7 times faster than baseline CSR. We also evaluated the "Embedded Julius" on a large vocabulary task (20,000 words). It shows almost real-time processing (1.25 x RT). Finally, We introduced Q & A guidance systems developed for embedded applications