|M.Sc Student||Altman Andres|
|Subject||Characterization of Voice and Speech during Imitation|
|Department||Department of Biomedical Engineering||Supervisor||Professor Emeritus Isak Gath|
The present research tries to define and quantify vocal and speech features that are modified during imitation and imposture. The assumption is that during imitation the imitator changes his voice in a similar manner as during imposture but to a greater degree. Thus, by analyzing the acoustic features in the imitator's speech it would be possible to get clues as to the features that are being used by impostors and as to the physiological changes that are involved in this process. The database used was based on recordings from one professional imitator. Two speakers with perceptually grossly different voices were chosen to serve as the target voices for the imitator. In the macro level, the pitch contour was calculated. The distance between the speaker’s pitch contour and the imitation/imposture pitch contour was calculated using Dynamic Time Warping (DTW). It was found that the DTW distance was a good indicator for the quality of the imitation/imposture. In the micro level the mean pitch for short segments of the phonemes /a/ and /e/ was estimated together with the first four formants and the cross-sectional areas of the vocal tract. It was found that the imitator was able to change his pitch in a wide range, but still was limited by his specific anatomy when trying to change the pitch in access of 23%. No clear relation was found between the formants and the imitator’s ability to imitate the speaker. The imitator modified his vocal tract cross-sectional areas following similar modifications in the speaker cross-sectional areas. The cross-sectional areas that were modified corresponded to specific anatomical structures involved in the pronunciation of the different phonemes.