Single-Channel Speech Separation Based on Deep Clustering with Local Optimization