基于非均匀矢量量化的孤立数字语音识别

本文讨论了基于非均匀矢量量化、隐马尔可夫模型(HMM)的孤立数字语音识别系统。在现有的连续密度隐马尔可夫模型多说话人孤立数字识别系统中,通常采用 LBG 算法建立矢量码本,并采用全搜索识别算法,这样的结果限制了识别精度和识别速度。本文提出了一种新的系统算法,即用非均匀矢量量化(Non-Uniform Vector Quantization——NUVQ)取代原矢量量化部份,实验结果证明,本系统在识别速度和识别精度上都有了较大的改善。

Speech Recognition of Isolated Digits Based on Non-Uniform Vector Quantization

Speech recognition of isolated digits based on NUVQ is discussed in this paper. LBG algorithm and full search algorithm are usually adopted in some system whi- ch is being researched, such as some recognition system for many speakers, but speech and quality of recognition is limited. A new method, using Non-Uniform Vector Quaniization (NUVQ), is presented in this paper. The experimenec shows that the system obtains significantly improvement in speech ,and quality of reco- gnition.