AAC窗型判定和量化模块的改进算法

An Improved Algorithm for AAC Window Determination and Quantization

  • 摘要: 由于先进音频编码(AAC, Advanced Audio Coding)标准中使用了较为复杂的心理声学模型和两层嵌套循环量化模块,使得其编码复杂度明显增加,从而影响了实时编码,所以,对于有实时编码要求的系统,必须对标准使用的心理声学模型和量化模块进行改进。本文针对AAC心理声学模型中的窗型判定子模块,提出了基于时域方差变化率的窗型判定算法;并且通过对AAC量化噪声评估模型的推导,设定量化因子初值,在可变比特率条件下无需循环就可实现量化编码,在固定比特率条件下可单循环实现量化编码且在96kbps码率条件下量化循环次数为1-2次。改进后的AAC编码器在TMS320C6727B实验平台上窗型判定模块运算量为1.79每秒百万时钟周期(MCPS, Million Cycles Per Second),量化模块为14.67 MCPS,并且经测试证明音质并无明显失真。

     

    Abstract: In order to achieve a better coding performance, Advanced Audio Coding (AAC) uses the complex psychoacoustic model and the two nested-loop quantization, but the high-level complexity of the above algorithms will badly affect the real-time encoding, so it is necessary for the real-time encoding system to improve the algorithm for AAC window determination and quantization. For the window determination in AAC psychoacoustics model, this paper proposes a new window determination method which is based on the variance ramp in time-domain, on the other hand, in the quantization of this paper, no loop is required for the variable bit rate condition and only single loop is required for the constants bit rate condition, the quantization algorithms is realized by the deriving of the accurate model of quantization noise and the defining of the initial scale-factor in the quantization. Moreover, for the constants bit rate condition, the number of quantization iterations at 96kbps bit rate is only about 1-2. In the last examination, the test results showed that our window determination algorithms requires only 1.79 MCPS and our quantization algorithms needs only 14.67 MCPS in TMS320C6727B platform , while the encoding audio quality have almost no impairment.

     

/

返回文章
返回