next up previous
次へ: Experimental results 上へ: Model Composition by Lagrange 戻る: Polynomial-approximation

Analysis of the approximation

To analyze the accuracy of the approximation, and to compare it with other methods, a set of one-dimensional vectors was generated for speech parameter by Monte-Carlo simulation. These speech vectors were corrupted by adding noise at different SNR. Noise vectors were also generated by Monte-Carlo simulation.

The means of corrupted speech estimated by different methods have been plotted in Figure 3. The result shows that the Lagrange polynomial approximation (LPA) outperforms VTS-1 and Log-max approximations. The mean estimate given by Lagrange polynomial approximation is almost same as given by Monte-Carlo simulation, however cutting down the computational cost to large extent.