Watermark-Aided Pre-Echo Reduction in Low Bit-Rate Audio Coding

摘要

In the context of low bit-rate audio coding, a system for pre-echo reduction, aided by data embedding, is presented. The proposed system aims at correcting the temporal envelope of the decoded signal using side information. The restored signal is reconstructed at the receiving end from two components: the decoded signal, generated by the core decoder, and the temporal envelope of the original signal, represented by a few parameters that are transmitted as embedded data in the audio signal. The audio signals affected by the attack smoothing distortion, concerned by the pre-echo reduction system, are of percussive type such as Castanet, cymbal, and arabic sounds. The data describing the envelope is transparently embedded at a rate not exceeding 50 bps and with a bit error detection of approximately 1.4 .10~4. The considered watermarking system consists in an additive time domain technique in which the spectral-shaped watermark signal is generated using an auditory masking model. The performance of the proposed system is evaluated for single and multiple coding/decoding processes performed by an audio MPEG compression (AAC and MP3). Performance evaluation is done through objective perceptual measures provided by the PEMO-Q software. The experimental results exhibit an efficient reduction of the pre-echo and a significant improvement of the audio quality, especially for the MP3 codec.