基于非均匀谱压缩特征的模型补偿新算法.pdfVIP

Vo 1.29No.6 第 29 卷第 6 期 电子与信息学报 2007 年6 月 Journal of Electronics Information Technology Jun. 2007 基于非均匀谱压缩特征的模型补偿新算法 宁史新 韦岗 孔祥祝 (华南理工大学电子与信息学院 广州 510640) 摘 要z 在信噪比依赖的非均匀谱压缩(SNSC)鲁棒语音特征提取技术和 VTS 算法的基础上，该文提出了一种新的 MC-SNSC 模型补偿算法。SNSC 技术是一种根据人类昕觉对声音强度-响度感知转化关系的谱幅度变化操作和噪声 抑制技术。基于对数谱域的噪声以及 SNSC 特征提取对语音信号特征所产生的失配函数，推导出了 MC-SNSC 模 型补偿算法。实验证明使用这一新算法，识别率比当前较理想的 VTS 和 PMC 算法有很明显的提升，算法的复杂 度较VTS 等算法仅有轻微的增加。 关键词g 语音识别:模型补偿;非均匀谱压缩 中固分类号: TN912.3 文献标识码:A 文章编号: 1009-5896(2007)06-138~05 Novel Model Compensation ßased on Non-uniform Spectral Compression Features Ning Geng-xin Wei Gang Kong Xiang-zhu (School of Electronic lnformαtion Engineering, South China University of Technology, Guangzhou 510640 , China) Abstract: A novel model compensation method is proposed , which integrates the Vector Taylor Series (VTS) approach with a robust feature extraction technique called SNR-dependent Non-uniform Spectral Compression (SNSC). The SNSC method is a spectral operation of magnitude transformation which resembles the human intensity-to-loudness conversion proc创s and de-emphasizes noisy bands. Based on this mismatch function , which models the effect of the noise onto the clean speech in the Log-spectral domain together with the SNSC , a new rnodel cornpensation procedure is derived. By adopting this novel rnodel cornpensation approach , significant irnprovernent over the PMC and VTS rnethod can be found in different additive noisy environrnents at the expense of slight increase in cornputat