Structured Gaussian Mixture model based product VQ
2010 (English)In: 18th European Signal Processing Conference (EUSIPCO-2010), EUROPEAN ASSOC SIGNAL SPEECH & IMAGE PROCESSING-EURASIP , 2010, 771-775 p.Conference paper (Refereed)
In this paper, the Gaussian mixture model (GMM) based parametric framework is used to design a product vector quantization (PVQ) method that provides rate-distortion (R/D) performance optimality and bitrate scalability. We use a GMM consisting of a large number of Gaussian mixtures and invoke a block isotropic structure on the covariance matrices of the Gaussian mixtures. Using such a structured GMM, we design an optimum and bitrate scalable PVQ, namely an split (SVQ), for each Gaussian mixture. The use of an SVQ allows for a trade-off between complexity and R/D performance that spans the two extreme limits provided by an optimum scalar quantizer and an unconstrained vector quantizer. The efficacy of the new GMM based PVQ (GMPVQ) method is demonstrated for the application of speech spectrum quantization.
Place, publisher, year, edition, pages
EUROPEAN ASSOC SIGNAL SPEECH & IMAGE PROCESSING-EURASIP , 2010. 771-775 p.
, European Signal Processing Conference, ISSN 2076-1465
speech coding, vector quantization
Research subject SRA - ICT
IdentifiersURN: urn:nbn:se:kth:diva-46587ISI: 000349999100156ScopusID: 2-s2.0-84863799825OAI: oai:DiVA.org:kth-46587DiVA: diva2:453915
18th European Signal Processing Conference, EUSIPCO 2010; Aalborg; Denmark; 23 August 2010 through 27 August 2010
QC 201111042011-11-032011-11-032015-12-07Bibliographically approved