Reduced Complexity Tree Coding of Speech in the Presence of Background Impairments

点击翻译按钮获取中文摘要

Extraordinary advances in speech coding have been achieved in the past few years, with the ITU and digital cellular industries producing voice coding standards for telephony and videoconferencing with performance perhaps unimaginable even ten years ago. At rates below 12 kbits/s, however, these coders are sometimes criticized for their implementation complexity and for their difficulty in coding speech in the presence of extraneous background impairments. The complexity issue is omnipresent, but becomes increasingly important in the mobile communications environment, both in digital cellular and emerging personal communications systems, where battery power is a dominating issue. Furthermore, there are many applications being developed and proposed for packet-switched videoconferencing at various rates, where it is desirable to have voice coders implemented either wholly in software on standard platforms (Macs, PCs, Sparcs) or whose implementations do not usurp large portions of the capabilities of a DSP chip. Being able to code speech in the presence of background impairments has always been of interest, but has recently moved to the fore because of the mobile environments encountered in digital cellular mobile radio and evolving personal communications systems. Desktop applications also present a challenge for coding with background impairments, since hands-free interaction is critical and thus even some relatively quiet office environments do not have the handset to aid in rejecting the background noise. The challenge is therefore to achieve communications quality speech at 8 kbits/s with medium-to-low complexity for a variety of implementation approaches without changing background impairments into annoying artifacts. This research approaches this problem by starting with a low delay waveform following coder, called a tree coder, removing the low delay requirement and employing frequency domain based variable rate techniques, forward adaptation ideas, code tr ee adaptivity innovations, and improved perceptual weighting to reduce implementation complexity while retaining reconstructed speech quality.