ISCA Archive Eurospeech 1995
ISCA Archive Eurospeech 1995

Analysis of acoustic features affecting speaker identification

Norio Higuchi, Makoto Hashimoto

In order to investigate the influence of individual acoustic features in the human process of speaker identification, the authors propose a new model to predict the perceptual contribution rate of each acoustic feature. The perceptual contribution rate is measured by a hearing test using speech resynthesized by swapping the acoustic features of natural speech, while its predicted value is calculated using a model based on differences in the acoustic features, that is, a cepstral distance representing spectral information and a difference of the mean logarithmic voice fundamental frequency representing information of the voice fundamental frequency. It is demonstrated that contribution rates can be predicted with an error of 7.71% in RMS after optimization of the acoustic features' weighting factors.