| Paper ID | E-2-2.2 |
| Paper Title |
TV-CAR SPEECH ANALYSIS BASED ON THE L2-NORM REGULARIZATION IN THE TIME-DOMAIN AND FREQUENCY DOMAIN |
| Authors |
KEIICHI FUNAKI, University of the Ryukyus, Japan |
| Session |
E-2-2: Speech Analysis |
| Time | Wednesday, 09 December, 15:30 - 17:00 |
| Presentation Time: | Wednesday, 09 December, 15:45 - 16:00 Check your Time Zone |
|
All times are in New Zealand Time (UTC +13) |
| Topic |
Speech, Language, and Audio (SLA): |
| Abstract |
Linear Prediction (LP) is a mathematical operation for estimating an all-pole spectrum from the speech signal. It is an essential methodology in speech coding since LP coefficients viz., Auto-Regressive (AR) coefficients can be determined using a small amount of computation and quantized efficiently using Vector Quantization (VQ). Recently, l2-norm Regularized LP (RLP), and context-aware Time-Regularized LP (TRLP) analysis have been proposed and shown to improve performance. The former suppresses rapid spectral changes in the frequency domain, and the latter suppresses the rapid spectral changes in the time domain. In our previous study, we proposed the MMSE based Time-Varying Complex AR (TV-CAR) speech analysis that is the complex-valued and time-varying version of the LP and the RLP-based TV-CAR analysis. In this paper, we propose the novel l2-norm regularized TV-CAR analysis based on not only the TRLP but also the RLP and the objective evaluation using a F0 estimation applied with the estimated complex residual signals shows that the proposed method performs best. |