Errata
------
The paper "Fast, low-artifact speech synthesis considering global variance"
published at ICASSP 2013 contains the following errors:
- In section 2.1 the typical value for $\omega$ should be $3 T$ instead of
$1 / (3 T)$.
- In section 4.2 the formula for the LSPA weight $w_t(\lambda)$ only applies
for $\lambda > 0$.
If $\lambda \leq 0$ then $w_t(\lambda)$ should be set to $1$.