Flexible spectrum coding in the 3GPP EVS codec

Multi-frame Coding of LSF Parameters Using Block-Constrained Trellis Coded Vector Quantization

“…The discussed trellis structures are based on a rate-1/2 convolutional code, which has N=2 v trellis states and two branches entering/leaving each trellis state [14][15][16][17][18][19]. For any 0 sv , a block-constrained trellis structure that allows 2 s initial trellis states and exactly 2 v-s terminal trellis states for each allowed initial trellis state.…”

“…Given a block of m source scalars or vectors, a single Viterbi algorithm encoding [20], starting from the allowed initial trellis states, proceeds in the normal trellis search up to the stage m-s. A unique terminating path, possibly dependent on the initial trellis state, is prespecified for each trellis state at stage m-s through stage m. For the BC-TCQ of LSF parameters, each trellis stage corresponds to one dimension element of the LSF vector [16]. As to the BC-TCVQ of LSF parameters for wideband speech, one trellis stage is associated with two neighboring elements of the LSF parameters [17][18][19]. The LSF parameters show the natural ordering property in the same frame which indicates their intra-frame correlation dependencies [1,21].…”

“…The LSF parameters show the natural ordering property in the same frame which indicates their intra-frame correlation dependencies [1,21]. The associated one dimension element or two neighboring elements in each stage maintain the ordering property, which can be exploited to design the intra-frame predictive BC-TCQ and BC-TCVQ, respectively [16][17][18][19].…”

“…Blockconstrained TCQ (BC-TCQ) requires exactly one bit per source sample to specify the trellis path with low complexity [16]. Block-constrained TCVQ (BC-TCVQ) of LSF parameters for wideband speech coders was also proposed by combining TCVQ with BC-TCQ [17][18][19].…”

“…The existing trellis encoding methods for LSF parameters [16][17][18][19] are single-frame quantization schemes. In this paper, the efficient BC-TCVQ is developed for multi-frame LSF parameters quantization.…”