Text this: Principal component analysis for HMM-based pathological voice detection