先进音频编码中的非均匀量化器的动态调整研究

A study on dynamic adjustment of the non-uniform quantizer in advanced audio coding

非均匀量化器广泛运用在感知音频编码器中.通过对非均匀量化器失真的研究,引入了量化器的群能量失真概念.该概念定义为量化前的原始信号能量和量化后的重建信号能量的误差期望值.根据此概念提出了零群能量失真量化器,这种量化器能使原始信号能量和重建信号能量保持守恒.在量化器的量化电平间使用零群能量失真原则,约束量化电平间的群能量失真,可以获得与原始量化谱分布相关的动态量化区间划分.通过对原始量化谱的分布进行频数统计,近似计算动态量化区间划分,修改量化器的取整方式,使用该近似的动态量化区间划分,实现了随输入信号分布变化的动态调整量化器.客观音质评价实验显示,使用动态调整量化器先进音频编码,与使用标准推荐的量化器相比,在相同比特率下,客观音质评价指标失真指数和噪声掩模比都有所提高,编码器音质得到改进,且音质改进程度随着比特率的增加而增大.而基于对比听音实验的主观音质退化程度比较表明,使用动态调整量化器的先进音频编码,与标准算法相比,在相同码率下有更高的音质还原度;在维持标准算法音质水平的前提下,可节省一定比特率.该动态调整量化器不改变音频编码器自身结构,在不显著增加的计算量和存储量的条件下,提高了编码器性能,且适用于多种类型音频编码器.

Non-uniform quantier is widely used in perceptual audio coders. In existing studies, quantization distortion is measured by mean square error. In this paper, through the study of non-uniform quantier, the cluster energy distortion is introduced. The cluster energy distortion is defined as the mathematical expectation of the energy error between the original signal before quantization and the reconstructed signal after quantization. With the cluster energy distortion principal, a quantizer without cluster energy distortion is proposed. This quantizer keeps the energy of the original signal and the reconstructed signal conservation. By applying the zero cluster energy error distortion rule among quantization levels to constraint the cluster energy distortion between quantization levels, a partition method of the dynamic quantization threshold can be obtained. This partition method correlates with the distribution of the raw quantization spectrum. By counting the appearance frequency in small divided spaces to obtain the approximate distribution of the raw quantization spectrum, an approximate solution of the dynamic quantization threshold could be calculated. With this dynamic quantization threshold by modifying the rounding operation of quantizer, a dynamic adjustment quantizer which follows the distribution of input signal is designed. This dynamic adjustment quantizer is applied in the advanced audio coding. Objective audio quality evaluation based on the perceptual evaluation of audio quality method shows that the encoder with the proposed dynamic adjustment quantizer has a better encoding performance than the encoder with the recommended quantizer that is defined in the advanced audio coding standard. The distortion index and the noise to mask radio, which are key objective audio quality measures are improved. This improvement becomes more significant as the bitrate increases. Subjective audio quality degeneration evaluation based on hearing test also shows that, at 218 kbsp bitrate, the encoder with the dynamic adjustment quantizer has less audio quality degeneration than the encoder with the recommended quantizer. With about 3 ~ reduction of the bitrate, the encoder with the dynamic adjustment quantizer keeps the same auditory perception level as the eneoder with the recommended quantizer. The dynamic adjustment quantizer is independent of the structure of the encoder, which only costs a slight increasing of computational complexity and storage space. This proposed method could also be applied to other encoders.