One of the most noticeable (and difficult to remove) artifacts in low bit-rate audio codecs is the high-frequency amplitude modulations, so-called 'birdies' (or 'warbling' artifacts). In this paper, we investigate the theoretical reasons why such artifacts occur, and in particular we show that they cannot be avoided with complete representations, such as critically sampled subband schemes, under a simple thresholding / quantization operation. In the case of the Modified Discrete Cosine Transform (MDCT), that forms the basis of many current coders, it is possible to compute explicitly the time-dependency of the coefficients of a pure sinusoid. From this, we derive a simple pre-filtering algorithm, that suppresses most of the frequency lines prone to warbling.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.