A Hybrid Approach to Electrolaryngeal Speech Enhancement Based on Spectral Subtraction and Statistical Voice Conversion