SYSTEM AND METHOD FOR AUTOMATICALLY CLASSIFYING THE AFFECTIVE CONTENT OF SPEECH

摘要

The classification (18) of the emotional content of speech (10) employs (16) acoustic measures (14) as classification input: pitch, spectral envelope shape (25) as either mel-frequency cepstral coefficients (MFCC) or linear prediction coefficients (LPC), variances in pitch and envelope, pitch magnitude, and energy. Changes in the spectral shape of the speech signal may distinguish long, smoothly varying sounds from quickly changing sound, which may indicate (18) the emotional state of the speaker. Different variations of pitch and spectral shape coefficients can be measured and analyzed to assist in the classification of individual utterances. Analysis can be performed on an entire utterance (28d), multiple segments of an utterance (10), or first, middle, and last thirds (28a, b, c) of an utterance.