摘要
研究了噪声背景条件下的语音特征提取。使用特征补偿的方法,并运用听觉心理学的三个概念:临界带谱分析、等响度曲线、强度响度听觉幂率,实现了PLP语音特征参数的提取。在此基础上用RATSA技术和半升正弦函数倒谱提升技术来改进此参数,并获得了良好的效果。
A feature extraction method is studied in this paper. PLP parameter is extracted using feature compensation method and three concepts of auditory psychology including critical- band spectral resolution, equal-loudness preemphasis and intensity-loudness power law of hearing. Then, on this basis the parameter is improved by using RASTA technique and Half Raisedsine Function Cepstral Exaltation, good performance for speech feature extraction under high noise is obtained.
出处
《太原理工大学学报》
CAS
北大核心
2009年第3期222-224,共3页
Journal of Taiyuan University of Technology
基金
国家自然科学基金(60472094)
山西省自然科学基金(20051039)
关键词
特征提取
听觉模型
感觉加权线性预测
半升正弦函数
feature extraction
auditory model
perceptual linear prediction (PLP)
half raised-sine function (HRSF)