基于频域线性预测心理声学掩蔽模型的音频编解码器

Audio Codec Using Frequency Masking Based on Frequency Domain Linear Prediction

频域线性预测给出了信号的希尔伯特包络的近似。基于频域线性预测的编解码器运用长时分割,很好地保持了时域包络信息。该编解码器能够重建高质量的信号,但是编码效率不高。将频域掩蔽引入到时域线性预测编解码器用以减少比特率。频域掩蔽是一个听音现象,如果另一个强度较大的声音出现,关注声音的听音阈值将增加。心理声学模型用于估计频域线性预测载波信号的听力阈值和绝对听力阈值。频域子带频域线性预测载波信号的比特配置根据听力阈值和绝对听力阈值计算得到。应用频率掩蔽,比特率下降5%。该文方法的效果应用音频质量感知评价和MUSHRA方法进行了测试。

Frequency Domain Linear Prediction （FDLP） gives an approximation of the Hilbert envelopes of a signal,which has been proved to contain most of the speech information.FDLP based Codec works with long temporal segments and keeps the information carried by the time-domain envelopes very wel .The codec gives good quality of the reconstructed signal,but is not efficient enough.This paper introduces Frequency masking to FDLP based codec to reduce the bit-rate.Frequency masking is a hearing phenomenon that the hearing threshold of a sound wil increase if an intense sound exists simultane-ously.The psychoacoustics model is used to estimate the hearing threshold and the absolute threshold of hearing （ATH） of the FDLP carrier signals,and bit al ocation for frequency sub-bands FDLP carrier signal is calculated according to the threshold and ATH.6% bit-rate reduction is obtained with the application of the frequency masking.