Speaker recognition has received a great deal of attention from the speech community, and significant gains in robustness and accuracy have been obtained over the past decade. However, the features used for identification are still primarily representations of overall spectral characteristics, and thus the models are primarily phonetic in nature, differentiating speakers based on overall pronunciation patterns. This creates difficulties in terms of the amount of enrollment data and complexity of the models required to cover the phonetic space, especially in tasks such as identification where enrollment and testing data may not have similar phonetic coverage. This dissertation introduces new features based on vocal source characteristics intended to capture physiological information related to the laryngeal excitation energy of a speaker. These features, including RPCC, GLFCC and TPCC, represent the unique characteristics of speech production not represented in current state-of-the-art speaker identification systems. The proposed features are evaluated through three experimental paradigms including cross-lingual speaker identification, cross song-type avian speaker identification and mono-lingual speaker identification. The experimental results show that the proposed features provide information about speaker characteristics that is significantly different in nature from the phonetically-focused information present in traditional spectral features. The incorporation of the proposed glottal source features offers significant overall improvement to the robustness and accuracy of speaker identification tasks.
展开▼
