A Framework for Predicting Speech Quality Using Detectability of Multiple Distortions