CAJ | 학술논문

为了提高深度模型的编码重构性能,本文为传统对比散度(Contrastive divergence,CD)添加了基于交叉熵的重构误差约束。利用改进后的算法训练了重构性深度自编码机(Reconstructive deep au-to-encoder,RDAE),并用 RDAE 替换混合激励线性预测编码(Mixed excitation linear prediction, MELP)语音编码器中 LSF 参数的矢量量化方法。测试结果表明,改进后的算法在损失一定模型似然度的条件下获得了重构性能的提升,当 RDAE 隐藏层结点设为19 bit 时,本文方法所测得的加权 LSF 距离、重构语音质量、谱失真指标在训练集和测试集上均优于25 bit 矢量量化方法,即利用本文方法改进的 MELP 编码器,在不降低语音质量的条件下,可将 MELP 编码速率从2.4 kb/s 降低至2.1 kb/s,编码速率降低了12.5%。
위료제고심도모형적편마중구성능,본문위전통대비산도(Contrastive divergence,CD)첨가료기우교차적적중구오차약속。이용개진후적산법훈련료중구성심도자편마궤(Reconstructive deep au-to-encoder,RDAE),병용 RDAE 체환혼합격려선성예측편마(Mixed excitation linear prediction, MELP)어음편마기중 LSF 삼수적시량양화방법。측시결과표명,개진후적산법재손실일정모형사연도적조건하획득료중구성능적제승,당 RDAE 은장층결점설위19 bit 시,본문방법소측득적가권 LSF 거리、중구어음질량、보실진지표재훈련집화측시집상균우우25 bit 시량양화방법,즉이용본문방법개진적 MELP 편마기,재불강저어음질량적조건하,가장 MELP 편마속솔종2.4 kb/s 강저지2.1 kb/s,편마속솔강저료12.5%。
In order to improve the reconstruction performance of deep models,reconstruction error con-straint based on cross entropy is added to traditional contrastive divergence (CD)algorithm.The im-proved algorithm is used to train reconstructive deep auto-encoder(RDAE),which is used to replace the vector quantization method for LSF in MELP speech coding algorithm.Experimental results show that the improved CD algorithm improves the deep model gain reconstruction performance while costing some likelihood of the model.When the node number of the hidden layer of RDAE is set to 1 9 bit,the indica-tors,which include the weighted LSF distance,the performance of reconstructed speech,and the spec-trum distortion,perform better in both training set and testing set by the proposed method than by the vector quantization method at 25 bit.That is to say,the coding bitrate of the MELP coder is reduced from 2.5 kb/s to 2.1 kb/s.The reduction rate of the coding bitrate is up to 12.5%,while the speech quality remains.