next up previous
Next: Rezayee's and Gazor's noise Up: Speech Enhancement Summaries Previous: Minimum-controlled recursive averaging noise

Projection error measure

Let $v_{ij}$ be the $j^{th}$ eigenvector of the $i^{th}$ frame. The noise subspace is given by the span of the noise eigenvectors, $v_{\mathcal{N}k}$:

\begin{displaymath}
\mathcal{S_N}=span\{v_{\mathcal{N}k}\vert k=1,\ldots,D\}
\end{displaymath}

The projection error measure for the $i^{th}$ frame is given by

\begin{displaymath}
\begin{array}{lll}
& & \frac{1}{D} \sum_{j=1}^{D} \Vert v_...
...l{N}}V_{\mathcal{N}}')V_{i} \Vert _{columns} \\
\end{array}
\end{displaymath}



Vinesh Bhunjun 2004-09-17