"Scalable Audio Compression for Multimedia Applications"

点击翻译按钮获取中文摘要

Scalable signal compression algorithms are required by the evolving communication network, which is composed of a variety of channels with widely differing static and dynamic capacities, and by the recent trend toward incremental capacity and bandwidth reservation channels. Current multimedia compression algorithms implicitly assume simple point to point transmission as well as fixed bandwidth allocation, and result in under-utilization of the network resources. The development of truly scalable coders with rate controlled by the local source statistics, network traffic load, and channel condition will directly impact the next generation of audio internetworking for multicasting, videotelephone, videoteleconferencing, and various applications for information retrieval from multimedia databases. This research focuses on new audio compression techniques for voice and music signals, that allow efficient scaling of bit rate and audio bandwidth within the framework of multimedia compression. The primary thrust centers on audio signal modeling with a sinusoidal signal representation, which is bandwidth and rate scalable and achieves good performance in a variety of coding environments from low-rate coding of narrow-band speech (2 kb/s) to high-quality coding of audio signals (16-64 kb/s). Methods under investigation include sophisticated phonetic and music classification to allow content-adaptive audio decomposition and modeling, generalized product code vector quantization, and in particular, multi- stage vector quantization, to implement scalable quantization techniques. Inter- stream optimization for joint compression of audio and video signals is studied with emphasis on potential applicability to packet transmission and in particular to TCP/IP Internet protocols. An important sub-problem is the inter-stream rate allocation optimization taking into account each stream's level of activity, user requirements, and sensitivity to quantization and channel errors. Sourc e-channel coding and error concealment issues are studied in order to maintain quality of service requirements under packet losses for both wired and wireless channels.