摘要
音频频带扩展是根据接收的宽带信号在解码端人为地重建出丢失的高频成分,以提升音频听觉质量。该文基于耳蜗滤波器倒谱参数提出了一种盲目式音频频带扩展方法。该方法模拟外耳听觉系统,提取耳蜗滤波器倒谱系数来描述宽带音频频谱信息,并利用Gauss混合模型对高频谱包络进行估计。结合基于最近邻匹配的谱细节恢复方法,实现了宽带向超宽带音频的有效扩展。主客观测试表明,该方法的重建音频质量优于基于传统音频特征的扩展方法。
Bandwidth extension of audio signals artificially restores the truncated high-frequency components from the transmitted wideband signal at the decoder to improve the auditory quality of the reproduced audio signals.This paper describes a blind bandwidth extension method based on the cochlear filter cepstral coefficients.The system emulates auditory peripheral hearing to calculate cochlear filter cepstral coefficients that more accurately describe the spectral information of the wideband audio.The Gaussian mixture model is used to estimate the spectral envelope of high frequencies.Finally,nearest neighbor mapping used to recover the fine structure at high frequencies is combined with the bandwidth extension of wideband audio signals to give superior wideband audio.Objective and subjective tests both indicate that the method achieves better performance than bandwidth extension methods based on conventional audio features.
出处
《清华大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2013年第6期913-916,共4页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金项目(61072089
60872027)
北京工业大学研究生科技基金项目(ykj-2012-7001)
关键词
音频信号处理
音频频带扩展
耳蜗滤波器倒谱系数
Gauss混合模型
最近邻匹配
audio signal processing
audio bandwidth extension
cochlear filter cepstral coefficient
Gaussian mixture model
nearest neighbor mapping