论文部分内容阅读
针对传统高斯混合模型在噪声环境下识别率明显下降的问题,在借鉴随机概率分布模型间的α因子融合机制基础上,提出基于可变因子α整合的高斯混合模型.该模型通过引入可变因子使得混合模型中不同成分所占的比重又得到一次调整.实验结果表明,通过对该模型参数进行重估计,在TIMIT/NTIMIT两种不同语料库和不同样本集的情况下识别率较传统高斯模型均有提高.尤其在噪声环境和α因子取最优值时,识别率可提高8%,在NIST评测数据集上与GMM-UBM系统对比,识别率也有提高.
Aiming at the problem that the recognition rate of traditional Gaussian mixture model obviously decreases in noisy environment, a Gaussian mixture model based on variable α integration is proposed based on a-factor fusion mechanism of stochastic probability distribution model. By introducing a variable factor Making the proportion of different components in the mixed model adjusted again.The experimental results show that the recognition rate of two different corpora and different sample sets of TIMIT / NTIMIT is higher than that of the traditional Gaussian model by re-estimating the model parameters Especially in noise environment and α factor, the recognition rate can be increased by 8%. Compared with GMM-UBM system, the recognition rate is also improved in the NIST evaluation data set.