Prediction and Optimization of Speech Intelligibility in Adverse Conditions