用于实时语声识别的连续VQ聚算法

A Continuous VQ Clustering Algorithm for Realtime Speech Recognition

本文阐述了一个用于实时语声处理的连续矢量量化聚算法(CVQC)。该算法把语声信息的时变性质用于训练和识别过程。与通常的动态时间弯曲(DTW)和矢量量化VQ方法比较,这个新算法的训练和识别速度均较快,码本较小,但仍保留两种方法的优点。这个新开发的算法着重于实时实现。此外,本文还介绍了一个以CVQC为基础的语声控制计算机命令输入系统。

This paper presents a continuous VQ clustering (CVQC) algorithm for realtime speech recognition, which incorporates the temporal information of speech into both training and recognition processes. In comparison with the conventional DTW and VQ methods, this new algorithm delivers faster training and recognition speed and smaller codebook size while still retains merits of both. Realt-time implementation is emphasized in the design of sophisticated algorithms. And a custom available voice controlled computer command input system based on CVQC is also introduced.