Advances in Similarity-Based Audio Compression

Existing lossy audio compression techniques such as MP3, WMA and Ogg Vorbis, for example, demonstrate great success in providing compression ratios which successfully reduce the data size from the original sampled audio. These techniques employ psychoacoustic models and traditional statistical coding techniques to achieve data reduction. However, these methods do not take into account the perceived content of the audio, which is often particularly relevant in musical audio. In this paper, we present our research and development work completed to date, in producing a system for audio analysis, which will consider and exploit the repetitive nature of audio and the similarities which frequently occur in audio recordings. We demonstrate the feasibility and scope of the analysis system and consider the techniques and challenges that are employed to achieve data reduction.