Prediction of Least Significant Bits from Upper Bits in Linearly Quantized Audio Waveform

Nishimura A. Prediction of Least Significant Bits from Upper Bits in Linearly Quantized Audio Waveform. In: AES Convention 146. Audio Engineering Society; 2019. Paper 10179. Available from: https://aes.org/publications/elibrary-page/?id=20312

@inproceedings{Nishimura2019_20312,
  author = {Nishimura, Akira},
  title = {{Prediction of Least Significant Bits from Upper Bits in Linearly Quantized Audio Waveform}},
  booktitle = {AES Convention 146},
  note = {Paper 10179},
  year = {2019},
  month = mar,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=20312}
}

TY  - CPAPER
TI  - Prediction of Least Significant Bits from Upper Bits in Linearly Quantized Audio Waveform
AU  - Nishimura, Akira
T2  - AES Convention 146
M1  - Paper 10179
PY  - 2019
DA  - 2019/03/06
UR  - https://aes.org/publications/elibrary-page/?id=20312
PB  - Audio Engineering Society
LA  - en
AB  - Bit-depth expansion of digital audio is essential for enhancing the quality of digital contents in re-mastering and up-conversion processes. The current study predicts the least significant bits for the bit-depth expansion from upper bits in linearly quantized samples of a framed audio waveform. A simulated annealing technique is applied to minimize the effective power of the residual signal derived from linear prediction of the framed waveform by localizing positions of the least significant bit (LSB) to be added in the frame. The results of computer simulation using various genres of 100 mono and 10-s music signals exhibit that the mean correct rate of the predicted LSB is 72% using 8-bit quantized waveforms. Measurements of the objective sound quality degradation reveal that the mean objective difference grade (ODG) of the 8-bit signals improved from -2.96 to -2.56 after addition of the predicted LSB.
ER  -