This paper explores robustness issues for real-time audio transmission over perturbed networks where multiple paths can be considered. Conjugate Vector Quantization (CVQ), a form of Multiple Description Coding, can improve the resilience to packet losses. This work presents a generalized CVQ structure, where K>2 different conjugate codebooks are trained to create the best resulting codebook. Experiments show that 4-description CVQ performs very closely to unconstrained VQ in clear channel conditions, while providing significant improvements in lossy channels. We also present a fast search algorithm which allows tradeoffs between computational complexity and memory storage at the encoder. This robust quantization scheme can encode sensitive information such as spectral coefficients in a speech coder or a perceptual audio coder.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.