中文摘要：

提出了一种流权值的优化方法。这种优化方法是基于似然比最大化准则和N-best算法。实验表明,这种新的方法即使在少量优化数据的条件下,也可以得到合适的流权值。而且,在不同的信噪比条件下,利用这种方法优化的多数据流隐马尔可夫模型,都可以有效、合理地融合音频和视频语音,提高语音识别系统的识别率。

英文摘要：

This paper proposed a novel stream-weight optimization method based on the likelihood-ration maximization criterion and the N-best algorithm. The proposed method had advantages that not only computational complexity was significantly reduced, but also audio-visual speech recognition performance was significantly improved by using a small optimization data set. Further experimental results demonstrate that the audio-visual speech recognition system provides significant enhancement of robustness in noisy environments.