基于GMM语音谱包络表示的编码算法研究

发布时间：2018-03-31 06:13

本文选题：语音编码　切入点：语音端点检测　出处：《南京师范大学》2017年硕士论文

【摘要】：低速率语音编码和极低码率语音编码是现代语音编码技术研究和发展的一个重要方向。对于语音编码技术来说,降低比特率是其不断发展的目标和动力。语音端点检测是语音编码过程中的重要一环,本文在研究了基础的基于方差的语音端点检测算法的基础上,提出基于ERB尺度子带划分的方差语音端点检测算法。实验结果表明,本文提出的基于ERB尺度划分的方差语音端点检测算法与基础的方差语音端点检测算法以及基于Bark尺度划分的方差语音端点检测算法相比,检测正确率,错检率,漏检率均有相应的改进。本文在深入研究基于高斯混合模型(Gaussian Mixture Model, GMM)对语音谱包络参数化表示的语音编码方法的基础上,提出一种分段编码改进算法。该算法将固定数量的语音帧划分为一个片段,在GMM对短时语音谱包络进行参数化的基础上,利用多项式拟合语音片段内的GMM参数,从而使得参数数量减少。仿真结果表明,本文算法码率对比基本的基于GMM的语音编码器有明显降低。另外本文还提出一种基于改进的高斯混合模型的甚低码率语音编码器。该编码器将多帧GMM参数组成一个超帧,对多帧参数联合编码,改进了 GMM模型,并用新模型对语音谱包络进行参数化表示。仿真结果表明,该编码器降低了编码速率,在0.86kb/s时仍然可以获得可接受的解码语音。
[Abstract]:Low rate speech coding and very low bit rate speech coding are an important direction in the research and development of modern speech coding technology. Speech endpoint detection is an important part of speech coding. In this paper, based on the research of the basic speech endpoint detection algorithm based on variance, the speech endpoint detection algorithm based on variance is studied. A speech endpoint detection algorithm based on ERB scale subband partition is proposed. The experimental results show that, Compared with the variance speech endpoint detection algorithm based on the ERB scale partition and the variance speech endpoint detection algorithm based on the Bark scale partition, the detection accuracy and error detection rate of the variance speech endpoint detection algorithm proposed in this paper are compared with those of the basic variance speech endpoint detection algorithm and the variance speech endpoint detection algorithm based on the Bark scale partition. In this paper, we study the speech coding method based on the Gao Si mixed model Gaussian Mixture Model (GMMM) for parameterized representation of the envelope of speech spectrum. An improved piecewise coding algorithm is proposed, in which a fixed number of speech frames are divided into a segment. On the basis of the GMM parameterization of the short-time speech spectrum envelope, the polynomial is used to fit the GMM parameters in the speech segment. Thus, the number of parameters is reduced. The simulation results show that, In addition, a very low bit-rate speech coder based on the improved Gao Si hybrid model is proposed. This encoder consists of multiple GMM parameters to form a superframe. The GMM model is improved for multi-frame parameter joint coding, and the new model is used to parameterize the speech spectrum envelope. The simulation results show that the encoder reduces the coding rate and can still obtain acceptable decoded speech in 0.86kb/s.
【学位授予单位】：南京师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3

【相似文献】