Quality enhancement of coded transient audio with a post-filter in frequency domain

摘要

This paper proposed a post-filter in frequency domain for enhancing the perceptual quality of the coded transient audio. The post-filter is used to filter the frequency coefficients of the coded transient and consists of a short-term post-filter which is derived from the Linear Predictive Coding (LPC) predictor based on the Discrete Cosine Transform (DCT) coefficients of the transient frame. In order to compensate the overall spectral tilt, a spectral tilt compensation filter is used following the short-term post-filter. As a result, the post-filter in frequency domain shapes the temporal noise in time domain while introducing a little distortion in audio. This kind of post-filter doesn't need side information from the encoder and can be used in any audio decoder.