ESCAPED-HUFFMAN AND ADAPTIVE RECURSIVE RICE CODING FOR LOSSLESS COMPRESSION OF THE MAPPED DOMAIN LINEAR PREDICTION RESIDUAL Noboru Harada 1 , Yutaka Kamamoto 1 , Takehiro Moriya 1 1 NTT Communication Science Laboratories, Nippon Telegraph and Telephone Corporation, Japan ABSTRACT ITU-T Recommendation G.711.0 has just been established. It defines a lossless and stateless compression for G.711 packet payloads (for both A-law and ȝ-law). This paper introduces some coding technologies proposed and applied to the G.711.0 codec, such as Plus-Minus zero mapping for the mapped domain linear predictive coding and escaped-Huffman coding combined with adaptive recursive Rice coding for lossless compression of the prediction residual. Performance test results for those coding tools are shown in comparison with the results for the conventional technology. The performance is measured based on the figure of merit (FoM), which is a function of the trade-off between compression performance and computational complexity. The proposed tools improve the compression performance by 0.16% in total while keeping the computational complexity of encoder/decoder pair low (about 1.0 WMOPS in average and 1.667 WMOPS in the worst-case). Index Terms— Speech coding, Standardization, ITU-T Recommendation G.711.0, Lossless compression of G.711 payload, Plus-Minus zero mapping, Escaped-Huffman coding, Adaptive recursive Rice coding, Mapped domain linear predictive coding. 1. INTRODUCTION The ITU Rec. G.711 [1] is widely used for narrowband telephony applications, including PSTN/GSTN and packet-based network applications such as VoIP, and has been used for many decades because of its proven voice quality, ubiquity, and utility. ITU has just established a lossless coding technology for G.711 encoded payloads. This new standard is ITU-T Rec. G.711.0 [2]. The G.711.0 codec may be used as a traditional codec and its use negotiated (end-to-end) by the end terminals (IP phones, conference bridge endpoints, etc.). Additionally, owing to its lossless and stateless design, G.711.0 may also be used as a lossless compression mechanism on any intermediate link (e.g., service provider VoIP backbone links at voice gateways) where G.711 is used by the end systems. G.711.0 employed in these transcoding applications provides bandwidth savings without any degradation of audio quality relative to G.711 since it is a lossless algorithm. For these gateway applications, low computational complexity is desired. The figure of merit (FoM), defined in the G.711.0 Terms of Reference (ToR) [3], was used to assess the tradeoff between complexity and signal compression during the design phase and in the G.711.0 selection process [4]. G.711.0 is a lossless compression algorithm that operates on 40, 80, 160, 240, and 320 samples per 8-kHz sampled G.711 input frame. The bit rate is variable and the size of the (compressed) output frame depends on the input signal characteristics. The minimum size of an encoded frame is one byte. The maximum size of an encoded frame is the input frame size plus one byte. Following coding tools are included in G.711.0: An uncompressed coding tool, constant coding tools (Constant Plus zero coding, Constant Minus zero coding, Constant non-zero coding), a Plus-Minus (PM) zero Rice coding tool, a binary coding tool, a pulse mode coding tool, a value-location coding tool, a fractional-bit coding tool, a min-max level coding tool, a direct linear predictive coding tool, and a mapped domain linear predictive (MDLP) coding tool. The MDLP coding is a kind of LP coding but especially designed for G.711 A-law and ȝ-law input (A similar scheme had been also proposed by F. Ghido, et. al. [5]). This paper introduces some new coding schemes proposed and applied to the G.711.0 codec, especially related to the MDLP coding tool. PM zero mapping is used to calculate the prediction residual and Escaped-Huffman (E-Huffman) coding combined with adaptive recursive Rice coding is used as an entropy coding scheme for the prediction residual. Test results are also examined in terms of the compression performance/computational complexity trade-off based on the FoM. Section 2 overviews the mapped domain linear prediction. Sections 3 and 4 introduce PM zero mapping and E-Huffman coding combined with adaptive recursive Rice coding, respectively. Section 4 shows the test results. Finally, Section 6 concludes the paper. 2. MAPPED DOMAIN LINEAR PREDICTION Figure 1 shows a block diagram of the MDLP encoding tool used in the G.711.0 encoder. The MDLP coding tool takes a sequence of N G.711 A-law, () A I n , or G.711 μ-law, () I n μ , symbols. First, these N G.711 symbols are converted into PCM () x n , 0 n N ≤ < in the uniform (linear) PCM domain and a short-term prediction is carried out in that domain using LP analysis. The prediction residual signal, however, is obtained in the range of [ 255, 255] − since the predicted value is subtracted from the target value int 8 () x n in the 8-bit logarithmic domain (denoted as the int8 domain in this paper). PARCOR coefficients are used to represent and signal the LPC parameter. Linear prediction is applied as follows: ( ) int8 PCM int8 int8 PCM int8 1 ˆ () ( ) P i i x n f a f x n i → → = ｧ ｷ ｨ ｸ = ⋅ − ｨ ｸ ｩ ｹ ｦ …(1) where i a is the i -th LPC coefficient of P -th order prediction, int8 ( ) x n i − and int8 ˆ () x n are the previous sample value and the predicted sample value in the int8 domain, and PCM int8 f → and 4646 978-1-4244-4296-6/10/$25.00 ©2010 IEEE ICASSP 2010