Encoding Quality

In order to evaluate the coding quality, ITU-R Recommendation BS. 1378 [26] is adopted as the objective audio quality measurement method, which defines the Objective Difference Grade (ODG). The ODG values ideally ranges from 0 to -4, where 0 corresponds to an imperceptible difference between reference and test signal and -4 corresponds to the very annoying difference. Therefore, the ODG value that is close to zero represents the better sound quality. Table 5.16 shows the scales of the ODG.

Table 5.16 The Scales of ODG

ODG scale Quality

0 Imperceptible -0.1 to -1 Perceptible but not annoying

-1.1 to -2 Slightly annoying

-2.1 to -3 Annoying

-3.1 to -4 Very annoying

Apart from ODG, Noise-to-Mask-Ratio (NMR) can be used as an alternative method to measure the sound quality. NMR is the ratio of the noise generated by the encoding process to the masking threshold calculated by the PAM. Negative NMR value represents the noise is masked by masking threshold. Therefore, the smaller NMR corresponds to the better sound quality. The values of ODG and NMR are calculated by EAQUAL [27] software. We compare the sound quality of the compressed audio with the sound of the uncompressed audio by

Table 5.17 and 5.18 show the results of the original 3GPP HE-AAC encoder and proposed HE-AAC encoder with bitrate at 48k and 32k bps. The decoder that we use is FAAD2 [32].

Table 5.17 The ODG and NMR of the Original and the Proposed HE-AAC Encoder of 48k bps Bitrate.

Original HE-AAC Pro posed HE-AAC

-8

Figure 5.19 The NMR of the original and the proposed HE-AAC of 48k bps bitrate.

Table 5.18 The ODG and NMR of the Original and the Proposed HE-AAC Encoder of 32k bps Bitrate.

-4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0

1 2 3 4 5 6 7 8 9

Track

ODG

Original HE-AAC Pro posed HE-AAC

Figure 5.20 The ODG of the original and the proposed HE-AAC of 32k bps bitrate.

-7 -6 -5 -4 -3 -2 -1 0

1 2 3 4 5 6 7 8 9

Track

NMR

Original HE-AAC Proposed HE-AAC

Figure 5.21 The NMR of the original and the proposed HE-AAC of 32k bps bitrate.

Chapter 6 Conclusion and Future Work

6.1 Conclusion

The main goal of this thesis is to accelerate the MPEG-4 HE-AAC encoder and to implement it on the TI C6416T DSP processor. Our proposed acceleration methods efficiently reduced the complexities of the HE-AAC encoder. These methods included transient detector acceleration, fast down-sampling filter, simplified block switching mechanism, low complexity psychoacoustic model, simplified TNS, fast quantization, and simplified window grouping module. The transient detector acceleration and fast down-sampling filter are the acceleration methods for the SBR portion. For the transient detector, we analyzed the compressed audio spectrum and the SBR frequency band tables in Section 3.3.3 . We observed that the frequency above 17k Hz is generally low power and is truncated at the end of process. Hence, in finding the transients, we simply calculated the signal energy of the frequency below 17k Hz to accelerate the transient detector. For down-sampling filter, we used the poly-phase decomposition to reduce its computations.

The simplified block switching, low-complexity psychoacoustic model, simplified TNS, fast quantization, and simplified window grouping module are the acceleration methods for the AAC portion. In the simplified block switching, we removed the high-pass filter to speed up the AAC encoder and the experimented data shows that removing the high-pass filter still maintained good audio quality. In the low complexity PAM approach, we reduced the calculation of spreading functions and spreaded energies by replacing them with a look-up table. For the simplified TNS, we used an early termination method to accelerate the TNS module. This method significantly reduced the computations of TNS. For fast quantization,

each scalefactor band. We also proposed a simplified step of checking the complete region of the scalefactor bands. For window grouping acceleration, we used only one group for the eight short windows to replace four groups.

Several experiments at different bitrates have been conducted to verify the acceleration methods. The speed-up performance, memory requirement, and audio quality were taken into consideration together. The speed improvement of the proposed HE-AAC was about 55 % over the original 3GPP HE-AAC under the same compiler optimization level. The computational complexity of the proposed HE-AAC encoder could be reduced to 50 MIPS.

For the memory requirement, the code size requirement was reduced by 39% and the RAM requirement was reduced by 10% when comparing to the 3GPP HE-AAC codec. As for the objective sound quality tests, we maintained the same level of the sound quality.

6.2 Future Works

Our MPEG-4 HE-AAC (aacPlus) codec is mainly concentrated on the aacPlus version 1.

The aacPlus v1 is a combination of AAC and SBR. SBR exploits the possibilities of a parameterized representation of the highband signals. However, aacPlus v2 adds a new technology to the aacPlus v1 in order to support lower bitrate coding. This new technology is called Parametric Stereo (PS). Figure 6.1 shows the aacPlus audio codec family.

AAC SBR Parametric

Stereo

+ +

aacPlus v1 aacPlus v2

aacPlus v2 = AAC + SBR + PS

Figure 6.1 aacPlus audio codec family

PS module increases the coding efficiency by exploiting a parametric representation of the stereo image of a given input signal. Then, in order to provide lower bitrate coding and to maintain a high audio quality, aacPlus v2 can be chosen as the audio codec. But the parametric stereo is also a time consuming module, and should be speeded up. Therefore, accelerating the aacPlus v2 and implementing it on the DSP system can be a useful and challenging task. Our acceleration methods for AAC and SBR can be a part of techniques used for accelerating the aacPlus v2.

References

[1] ISO/IEC JTC1/SC29/WG11, “Text of ISO/IEC 13818-7:2005 (MPEG-2 AAC 4^th edition)”, ISO/IEC JTC1/SC29/WG11 N7126, April 2005.

[2] ISO/IEC JTC1/SC29/WG11, “Draft ISO/IEC 14496-3:2001/Amd.2:2004 (Audio 3^rd Edition)”, ISO/IEC JTC1/SC29/WG11 N7027, Jan. 2005.

[3] T. Painter and A. Spanias, “Perceptual Coding of Digital Audio”, Proc. the IEEE, Vol. 88, Issue 4, pp. 451-515, Apr. 2000.

[4] M. Dietz and et al., “Spectral Band Replication, a novel approach in audio coding”, 112^th Audio Engineering Society (AES)Convention, Munich, May 2002.

[5] M. Wolters and et al., “A closer look into MPEG-4 High Efficiency AAC”, 115^th Audio Engineering Society (AES)Convention, New York, October 2003.

[6] P. Ekstrand, “Bandwidth Extension of Audio Signals by Spectral Band Replication”, IEEE Benelux Workshop on Model based Processing and Coding of Audio (MPCA-2002), Leuven, Belgium, Nov 15, 2002.

[7] M. Dietz and M. Wolters, “Enhancing Perceptual Audio Coding through Spectral Band Replication”, Coding Technologies, Nuremberg, Germany.

[8] Digital Radio Mondiale (DRM). Available: http://www.drm.org/ .

[9] Third Generation Partnership Project (3GPP). Available: http://www.3gpp.org/ . [10] Digital Video Broadcasting (DVB). Available: http://www.dvb.org/ .

[11] Coding Technologies. Available: http://www.codingtechnologies.com/ .

[12] 3GPP Telecommunication Standard 26.401: “General audio codec audio processing functions; Enhanced aacPlus general audio codec; General description”, Mar. 2005.

functions; Enhanced aacPlus general audio codec; Encoder specification; Advanced Audio Coding (AAC) part”, Sep. 2004.

[14] 3GPP Telecommunication Standard 26.404: “General audio codec audio processing functions; Enhanced aacPlus general audio codec; Encoder specification; Spectral Band Replication (SBR) part”, Sep. 2004.

[15] T.-H. Tsai, and et al., “Design of a low power psycho-acoustic model co-processor for MPEG-2/4 AAC LC stereo encoder”, Proc. of the IEEE Int. Symp. Circuits and Systems, Vol. 2, May, 2003, pp. 552-555.

[16] S.-W. Huang, and et al., “A low complexity design of psychoacousitc model for MPEG-2/4 advanced audio coding“, IEEE Trans. Consumer Elec., Nov., 2004.

[17] C.-Y. Lee and et al., “A fast audio bit allocation technique based on a linear R-D model”, IEEE Trans. Consumer Elec., Vol. 48, Issue 3, pp. 662-670, Aug. 2002.

[18] C.-M. Liu and et al., “A fast bit alloaction method for MPEG layer III”, Proc. of the ICCE, pp. 22-23, 1999.

[19] C.-T. Chen, “A new bit allocation for MPEG layer III,” Master thesis, Department of Computer Science, National Chiao Tung University, Hsinchu, Taiwan, ROC, 1998.

[20] Y.-L. Lin, “MPEG-2/4 low complexity AAC encoder optimization and implementation on a strongARM platform,” M.S. thesis, Department of Electrical and Control Engineering , National Chiao Tung University, Hsinchu, Taiwan, ROC, 2005.

[21] Texas Instruments, “TMS320C6000 Programmer’s Guide”, Literature number SPRU198F, Feb. 2001.

[22] Texas Instruments, “TMS320C6000 CPU and Instruction Set Reference Guide”, Literature number SPRU189F, Jan. 2000.

[23] Texas Instruments, “TMS320C64x Technical Overview”, Literature number SPRU395B,

[24] Sundance DSP System, “SMT 395 User Manual”, May 2001.

[25] Texas Instruments, “TMS320C6414T, TMS320C6415T, TMS320C6416T FIXED POINT DIGITAL SIGNAL PROCESSORS”, Literature number SPRU228H, NOV.

2003.

[26] ITU-R Recommendation BS.1387, “Method for objective measurements of perceived audio quality”, 2001.

[27] EAQUAL. [Online]. Available: http://www.mp3-tech.org/programmer/sources/eaqual.tgz [28] European Broadcasting Union (EBU), Sound Quality Assessment Material: Recordings

for Subjective Tests, Brussels, Belgium, Apr. 1988.

[29] M. Bosi and Richard E. Goldberg, Introduction to digital audio coding and standards.

Kluwer Academic Publisher Press, 2003.

[30] K. Sayood, Introduction to data compression. Morgan Kaufmann Publisher, 2000.

[31] FAAC. Freeware Advanced Audio Coder. v1.24 source code [Online].

Available: http://www.audiocoding.com/modules/mydownloads/

[32] FAAD2. Freeware Advanced Audio Decoder source code. [Online].

Available: http://www.audiocoding.com/modules/mydownloads/

[33] MPEG-4 AAC and High Efficiency AAC Reference Software ISO/IEC 14496-5 / FPDAM6: 2004. ISO/IEC JTC1/SC29/WG11 N6248, May 2004.

[34] H.-Y. Huang, “Design and Implementation of AAC Audio Coder,” M.S. thesis, Department of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan, ROC, 2004.

自 傳

黃育彰, 西元 1981 年生於高雄市。 西元 2004 畢業於台灣新竹國立交通大 學電機與控制工程學系，之後進入交通大學電子研究所攻讀碩士學位，於 2006 年取得碩士學位。研究方向為數位訊號處理、音訊壓縮。

Yu-Chang Huang was born in KaoHsiung in 1981. He received the BS degree in Electrical and Control Engineering from National Chiao Tung University (NCTU), HsinChu, Taiwan in 2004. He pursue master degree in Electronics Engineering at National Chiao Tung University. His research interests are digital signal processing and audio compression.