Lossless compression of audio data

数据压缩是使用比数据正常表示所需空间更少的空间来表示数据的过程。数据压缩可以是“无损的”，在这种情况下，原始数据可以完全恢复，或者是“有损的”，即只能恢复原始数据的近似值。无损数据压缩适用于文本或其他类型的数据，其中任何修改都会产生重大的有害影响。有损数据压缩经常用于图像或音频数据，其中退化程度根据某些“可接受性”概念进行调整。由于围绕MP3文件共享的知识产权问题，音乐的有损压缩目前受到了公众的关注。虽然有损压缩适用于诸如随意欣赏音乐之类的应用，但在其他情况下就不合适了。例如，从被动声纳系统获得的音频数据通常具有非常低的信噪比（SNR）；也就是说，期望的声音被嵌入到一个相对较大的环境噪声场中。使用有损压缩方案进一步降低信噪比可能是不可接受的。此外，有损压缩技术通常基于心理声学原理，即人类不会听到某些声音（例如，宽泛地说，在较大声的声音之前或之后不久出现的安静声音）；如果音频数据是由计算机软件处理的，这些原则是不合适的。本研究项目将研究音频数据的无损压缩，特别关注以下主题：(1)定义新的音频数据模型以优化可压缩性；(2)在有利的情况下，将这些模型区分为声纳数据、音乐和语音；(3)检查特定压缩方案的压缩工作与解压工作之间的权衡。

Data compression is the process of representing data using less space than that required by the data's normalrepresentation. Data compression can either be "lossless", in which case the original data can be recovered exactly, or "lossy", whereby only an approximation of the original data can be recovered. Lossless data compression is appropriate for textual or other kinds of data where any modifications would have significantdetrimental effects. Lossy data compression is frequently used for image or audio data, where the degree of degradation is adjusted according to some notion of "acceptability". Lossy compression of music is currently in the public eye because of intellectual property issues surrounding the sharing of MP3 files.While lossy compression is suitable for applications such as the casual enjoyment of music, in other situations it is not appropriate.  For example, audio data such as that obtained from passive sonar systems typically has a very low signal to noise ratio (SNR); that is, the desired sound is embedded in a relatively loud environmental noise field. Further degradation of the SNR by using a lossy compression scheme may be unacceptable. Also, lossy compression techniques are often based on psycho-acoustic principles which dictate that humans will not hear certain sounds (such as, loosely speaking, a quiet sound occurring shortly before or after a louder sound); these principles are inappropriate if the audio data is to be processed by computer software.This research program will study the lossless compression of audio data, with particular attention to the following topics: (1) defining new audio data models to optimize compressability; (2) differentiating thesemodels, if beneficial, towards sonar data, music and speech; and (3) examining the tradeoffs involved with compression effort vs. decompression effort for particular compression schemes.