论文部分内容阅读
为了提高感知线性预测系数(PLP)在噪声环境下的识别性能,使用子带能量偏差减的方法,提出了一种基于子带能量规整的感知线性预测系数(SPNPLP)。PLP有效地集中了语音中的有用信息,在安静环境下自动语音识别系统使用PLP可以取得良好的识别率;但是在噪声环境中其识别性能急剧下降。通过使用能量偏差减的方法对PLP的子带能量进行规整,抑制背景噪声激励,提出了SPNPLP,增强自动语音识别系统在噪声环境下的鲁棒性。在一个语法大小为501的孤立词识别任务和一个大词表连续语音识别任务上做了测试,SPNPLP在这两个任务上,与PLP相比,汉字识别精度分别绝对提升了11.26%和9.2%。实验结果表明SPNPLP比PLP具有更好的噪声鲁棒性。
In order to improve the performance of perceived linear prediction coefficients (PLP) in noise environment, a subband-based PSNR (SPNPLP) method is proposed based on subband energy deviation subtraction. PLP effectively concentrates the useful information in speech, and the automatic speech recognition system can obtain a good recognition rate using PLP in a quiet environment; however, its recognition performance drastically decreases in a noisy environment. The subband energy of PLP is regularized by using the method of energy deviation subtraction, and the background noise excitation is suppressed. SPNPLP is proposed to enhance the robustness of the automatic speech recognition system in noisy environments. In a task with a grammatical size of 501 isolated speech recognition task and a large vocabulary word continuous speech recognition task, SPNPLP performed an absolute improvement of 11.26% and 9.2% respectively on both tasks compared with PLP, . Experimental results show that SPNPLP has better noise robustness than PLP.