窄带语音带宽扩展算法研究
发布时间：2018-11-18 09:14
【摘要】：为了降低谱失真,提出了一种基于隐马尔科夫模型的窄带语音带宽扩展算法。首先,算法选取与宽带谱包络互信息大的参数构成特征矢量,并利用隐马尔可夫状态和过去观察特征矢量的联合先验概率估计条件后验概率。其次,以条件后验概率为基础,算法结合贝叶斯条件参数估计法和最小均方差准则估计宽带谱包络。针对宽带激励信号估计,基于信号高频和低频的谐波相关性,提出了一种中频激励扩展算法。实验结果表明,与传统的基于隐马尔可夫模型的带宽扩展算法相比,本文算法可降低0.187 dB的平均谱失真,将谱失真大于10 dB的语音帧减少了34.3%。
[Abstract]:In order to reduce spectral distortion, a narrow band speech bandwidth expansion algorithm based on hidden Markov model is proposed. Firstly, the parameters with large mutual information of wideband spectrum envelope are selected to form the feature vector, and the conditional posteriori probability of joint prior probability estimation of hidden Markov state and past observation feature vector is used. Secondly, based on conditional posteriori probability, the algorithm combines Bayesian conditional parameter estimation and minimum mean square error criterion to estimate wideband spectral envelope. For wideband excitation signal estimation, an intermediate frequency excitation expansion algorithm is proposed based on the correlation between high frequency and low frequency harmonics. The experimental results show that compared with the traditional bandwidth expansion algorithm based on hidden Markov model, the proposed algorithm can reduce the average spectral distortion of 0.187 dB and reduce the speech frames with spectral distortion greater than 10 dB by 34.3 points.
【作者单位】： 北京大学信息科学技术学院;深港产学研基地深圳市智能媒体和语音重点实验室;
【分类号】：TN912.3
[Abstract]:In order to reduce spectral distortion, a narrow band speech bandwidth expansion algorithm based on hidden Markov model is proposed. Firstly, the parameters with large mutual information of wideband spectrum envelope are selected to form the feature vector, and the conditional posteriori probability of joint prior probability estimation of hidden Markov state and past observation feature vector is used. Secondly, based on conditional posteriori probability, the algorithm combines Bayesian conditional parameter estimation and minimum mean square error criterion to estimate wideband spectral envelope. For wideband excitation signal estimation, an intermediate frequency excitation expansion algorithm is proposed based on the correlation between high frequency and low frequency harmonics. The experimental results show that compared with the traditional bandwidth expansion algorithm based on hidden Markov model, the proposed algorithm can reduce the average spectral distortion of 0.187 dB and reduce the speech frames with spectral distortion greater than 10 dB by 34.3 points.
【作者单位】： 北京大学信息科学技术学院;深港产学研基地深圳市智能媒体和语音重点实验室;
【分类号】：TN912.3
【参考文献】
相关期刊论文 前3条
1 郎s，
本文编号：2339596
本文链接：https://www.wllwen.com/kejilunwen/wltx/2339596.html
