Voice predicts affect during psychotherapy.

This study examined the relationship between emotions and the frequency and power characteristics of the voice in psychotherapy. The intensity of fear, anger, depression, and total affect in each of four interviews with one patient was rated every 20 seconds on a nine-point scale. Significant agreement among judges was achieved. Voice samples from each epoch in which there was sufficient speech were subjected to spectral analysis of the frequencies between 0 and 1000 HZ. These spectra were scored for nine frequency and power parameters. Multiple linear regression equations were then developed from two interviews, using the nine voice spectral variables as predictors and the mean ratings for each affect as the criterion variables. Significant multiple correlations were achieved between every rated affect and various combinations of voice variables. The beta weights and constants from these equations were then employed in the successful prediction of levels of anger, fear, depression, and total affect in one interview, and the levels of depression and total affect in another interview. In addition, epochs of conflict differed from "pure" affect epochs, and pure epochs of anger, fear, and depression differed from each other in various frequency and power characteristics of the voice. Voice spectral measures may be an objective means of identifying and quantifying affect in psychotherapy.