NMR Optimized Bitrate Transcoding for MPEG-2/4 AAC with LC Profile

• Main Authors: Te Hsueh Lai, 賴德宣
• Other Authors: Tihao Chiang
• Format: Others
• Language: en_US
• Online Access: http://ndltd.ncl.edu.tw/handle/27268286893252045774

碩士 === 國立交通大學 === 電子工程系所 === 94 === Real-time audio streaming services like music-on-demand (MOD), digital audio broadcasting (DAB), etc, deliver multimedia content over heterogeneous networks and to client devices with varying capabilities. To fit the network conditions and the clients’ capabilities, the bitrate adaptation based on the transcoding techniques is applied. We present a noise-to-masking-ratio (NMR) optimized MPEG-2/4 AAC LC transcoder, which is called as Fast Rate-Distortion Optimized Transcoder (FRDOT). In addition, FRDOT searches for the optimal scalefactor under the NMR criterion at a given bitrate. The computation of NMR difference is replaced by the derivation of signal-to-noise-ratio (SNR) difference since the audible masking thresholds of the input and output bitstreams are identical before and after transcoding. Within FRDOT transcoder, the SNR value is further converted to a noise-to-signal-ratio (NSR) to represent the distortion energy of audio signals. Therefore, the NMR optimized transcoding can be converted to the NSR optimized transcoding. The NSR optimized transcoding can find the optimal scalefactor increment according to the magnitudes of quantized input coefficients and the target bitrate. To speed up the search of optimal scalefactor increment, a table lookup technique is used. To further reduce the execution time, the bandwidth limiter is adopted to remove the iterative rate-distortion optimization of a frame. In addition, a bitrate control module is proposed to make the averaged bitrate of output bitstream close to the target bitrate. The experiment results show that the NMR value of FRDOT is better than the NMR value of cascaded transcoder (CT) by 0.5-3.0 dB at different bitrates and FRDOT can speed up CT by 5-8 times on the average.