A Multipulse Speech Coding Model via ℓ1/2-norm Minimization based Linear Prediction and Sparse Decomposition
- 10.2991/icaset-18.2018.14How to use a DOI?
- Multipulse speech coding, ℓ1/2-norm, Sparse linear prediction, Sparse decomposition.
Sparse linear predictive analysis using the ℓ1-norm minimization criterion has been shown to provide a valid alternative to the traditional linear predictive approach. The sparser the resulted speech residual is, the fewer pulses needed to represent it. To find a sparse residual further, we propose to use the ℓ1/2-norm as the optimization objective of linear prediction, and use an iteratively reweighted ℓ1 minimization approach to solve it. We also find that the procedure of determining the locations and amplitudes of the optimal pulses in the multipulse analysis is equivalent to a sparse decomposition problem, which is efficiently solved by the optimized orthogonal matching pursuit approach. The objective and informal subjective evaluations over the TIMIT database give proof of the effectiveness of this new model, performing better than the traditional approach for modeling and coding of speech.
- © 2018, the Authors. Published by Atlantis Press.
- Open Access
- This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).
Cite this article
TY - CONF AU - Suiqing Xue AU - Gang Min AU - Guolei Ren PY - 2018/04 DA - 2018/04 TI - A Multipulse Speech Coding Model via ℓ1/2-norm Minimization based Linear Prediction and Sparse Decomposition BT - Proceedings of the 2018 8th International Conference on Applied Science, Engineering and Technology (ICASET 2018) PB - Atlantis Press SP - 66 EP - 73 SN - 2352-5401 UR - https://doi.org/10.2991/icaset-18.2018.14 DO - 10.2991/icaset-18.2018.14 ID - Xue2018/04 ER -