基於Sphinx可快速個人化行動數字語音辨識系統

Quickly Personalizable Mobile Digit Speech Recognition System Based on Sphinx

本論文建立了一個透過網路提供數字語音辨識服務的系統，除了語音辨識功能也提供線上個人化調適功能來克服在不同環境中的噪音強健性。以英文數字辨識來說使只需要經過少許的調適就能夠在少許的時間內打造出正確率達80%以上的個人化英文數字語音辨識系統。Sphinx-4是專門為了研究而開發的工具，具有延展性、模組化、可插拔的架構，因為這些特性我們選擇使用Sphinx-4做為語音辨識系統的核心。為了讓選擇聲學模型與訓練語料及調適聲學模型上有一個依據，使用AURORA2語料庫訓練模型，台灣口音英語語料庫與Android裝置錄製的語料進行調適實驗，結果顯示使用EAT語者獨立的語料經100句調適後正確率能夠由80%進步到約90%；多環境模型經Android單人語料經25句能從平均70%提升到約95%的正確率。

In this paper, we introduce a system for on-line digit speech recognition services. Besides the speech recognition service in our system, we also provide adaptation function to improve the noise-robustness between di erent environment. In the case of English digit recognition, our recognition system can achieve over 80% accuracy for a specific speaker by using a few adaptation data. We use Sphinx-4 as a speech recognition kernel in our system. Because Sphinx-4 is a system prepared exclusively for researchers, it is a flexible, modular and pluggable framework. We provide our experiment results on AURORA2, EAT and Android device recording. We use AURORA2 database training models that adapt by EAT and Android device recording. The experimental results show we can get high accuracy after a few adaptation.