Speech feature modeling for robust stressed speech recognition

It is well known that the performance of speech recognition algorithms degrade in the presence of adverse environments where a speaker is under stress, emotion, or Lombard e ect. This study evaluates the e ectiveness of traditional features in recognition of speech under stress and formulates new features which are shown to improve stressed speech recognition. The focus is on formulating robust features which are less dependent on the speaking conditions rather than applying compensation or adaptation techniques. The stressed speaking styles considered are simulated angry and loud, Lombard e ect speech, and noisy actual stressed speech from the SUSAS database (available on CD-ROM through NATO RSG.10 research group, and soon LDC). In addition, this study investigates the immunity of LP and FFT power spectrum to the presence of stress. Our results show that unlike FFT's immunity to noise, the LP power spectrum is more effective than the FFT to stress as well as to a combination of a noisy and stressful environment. Two alternative frequency partitioning methods (M-MFCC, ExpoLog) are proposed and compared with traditional MFCC features for stressed speech recognition. It is shown that the alternate lterbank frequency partitions are more e ective for recognition of speech under both simulated and actual stressed conditions.