Robust model-based single-source localization in reverberant environments using a statistically enhanced TR-MUSIC framework