Low-cost smart transcoding algorithm between ITU-T G.729 (8 kbit/s) and 3GPP NB-AMR (12.2 kbit/s)

Networks interconnection causes interoperability problems between different speech coding formats. Today, tandem transcoding solutions (decoding/re-encoding) are generally used in communication chains. However intelligent transcoding solutions which exploit similarities between formats have been recently proposed in order to overcome tandem drawbacks (computational complexity, algorithmic delay, speech degradation). This paper gives an overview of these intelligent transcoding methods for CELP coders. Then, a low-cost intelligent transcoding algorithm between ITU-T G.729 (at 8 kbit/s) and 3GPP NB-AMR (at 12.2 kbit/s) is proposed. It is composed of four parts corresponding to the four CELP parameters conversions: LSP coefficients, fractional pitch lags, fixed codevectors and gains. A novel and computationally efficient method is described here for the fixed codevectors. The ACELP search is strongly focused on privileged positions, what considerably reduces the number of tested combinations. Objective and subjective quality tests show that this smart transcoding algorithm achieves a quality very close to that of the tandem while strongly reducing complexity with a shorter algorithmic delay.