Analysis and compensation of speech under stress and noise for environmental robustness in speech recognition