Abstract: Aiming at the problem that the Keyword Spotting(KWS) system can not calculate the state likelihood effectively. Based on nearest-neighbor approximation, a new technique named Feature Similarity of Adjacent Frames(FSAF) is proposed. Depending on the high similarity of adjacent frames, it uses some of the maximum mixtures of the previous frame to predict the maximum mixture used by the current frame to calculate the likelihood quickly. Experimental result shows the technique reduces the recognition time by 29.3% over baseline system, but the identification performance is kept, it can be applied to actual projects.

Key words: speech recognition, Keyword Spotting(KWS), GMM approximation, on-line garbage model, Feature Similarity of Adjacent Frames(FSAF), Hidden Markov Model(HMM)

摘要： 针对关键词检出系统中计算观察概率效率较低的问题，在最近邻近似方法的基础上，提出一种基于相邻帧特征相似性的方法。依据相邻帧之间的高相似性，利用产生前一帧特征矢量的若干个最大的混合分量，有效预测当前帧所使用的最大高斯混合分量，从而快速计算观察概率。实验结果表明，与基线系统相比，该方法在保持识别性能的前提下，识别时间可降低29.3%。

关键词: 语音识别, 关键词检出, GMM估算, 在线垃圾模型, 相邻帧特征相似性, 隐马尔科夫模型

CLC Number: 

• TP18