Mobile audio coding using lattice vector quantization based on Gaussian mixture model

In the paper, we describe a new lattice vector quantization algorithm based on Gaussian mixture model applied in mobile audio coding. In the algorithm, we uses Gaussian mixture model to classify the immittance spectral frequency parameters of the input signal to the divided M different Gaussian component according to probability density of different lattice vectors. And then for each Gaussian component, we will design a lattice vector to quantize the corresponding signal. Also, we adopt an optimal bit allocation algorithm based on the rate-distortion theory of the input signal. We demonstrate the efficiency of the proposed algorithm in the application of speech spectrum quantization in mobile audio coding. Experiments on encoding multiple audio test sequences show that the quantitative method that we present provides rather good performance and less storage space with moderate increase in computation complexity when compared with conventional quantization scheme used in AMR-WB+ audio codec.