摘要
为了解决人们在强噪声环境下,通过空气途径传递的语音信号会严重失真的问题,提出了一种基于深层双向长短期记忆-深度卷积神经网络(Deep Bidirectional Long and Short Term Memory-Deep Convolutional Neural Network,DBLSTM-DCNN)的骨导语音转气导语音的语音转换模型。该模型利用DBLSTM层收集和保存相邻连续帧的隐藏信息,再通过DCNN层来提取频域方面的特征信息,可以很好地解决由于骨导语音高频成份严重缺失导致的转换语音不够自然的问题。实验结果表明,该模型的语音质量感知评价(Perceptual Evaluation of Speech Quality,PESQ)、短时客观可懂度(Short-Time Objective Intelligibility,STOI)、对数谱距离(Log-spectral Distance,LSD)等客观评价指标均有良好的表现,证明了该模型在骨导语音转气导语音方面具有较好的转换效果。
In order to solve the problem that people's speech signals transmitted through air will be seriously distorted in strong noise environment,a speech conversion model from bone conduction voice to air conduction voice based on deep bidirectional long short term memory-deep convolutional neural network(DBLSTM-DCNN)is proposed in this paper.This model uses the DBLSTM layer to collect and save the hidden information of adjacent consecutive frames,and then uses the DCNN layer to extract feature information in the frequency domain.By this method,the problem that the converted voice is not natural enough due to the serious lack of high-frequency components of bone conduction voice can be well solved.The experimental results show that according to the good evaluation marks in objective indicators,such as perceptual evaluation of speech quality(PESQ),short-time objective intelligibility(STOI)and log-spectral dis-tance(LSD),this speech conversion model is confirmed to have good conversion effect from bone conduction voice to air conduction voice.
作者
储有亮
李梁
CHU Youliang;LI Liang(Chongqing University of Technology,Chongqing 400054,China)
出处
《声学技术》
CSCD
北大核心
2021年第6期815-821,共7页
Technical Acoustics
