Feature Extraction Of Speaker Identification Based On Phoneme F-ratio

In this paper, we employed the phoneme mean F-ratio method to investigate thedifferent contributions of different frequency region from the point of view of Chinesephoneme, and apply it for speaker identification. It is found that the speaker individualinformation depending on the phonemes is distributed in different frequency regionsof speech sound. Based on the contribution rate, we extracted the new features andcombined with GMM model. Compared with the MFCC feature, the identificationerror rate with the proposed feature was reduced by32.94%.Then, we conduct morphological analysis and acoustic modeling of the nasal andparanasal cavities to investigate the effects of the nasal cavity on speakercharacteristics. Morphological analysis showed that the nasal cavity possessesrelatively large variation across speakers. Acoustic effects results showed that theinter-speaker variation of the nasal tract affects spectra in the frequency range from2kHz to4kHz, which is in agreement with the results from our previous statisticalstudies.In speech production, the function of the velum is not a binary switch of on andoff. For the nasalized vowels and voiced stops in Japanese, the radiation probablymainly results from velum vibration. two mechanical experiments were conducted toreveal the acoustic incorporation of the transvelar coupling of the yielding velum.Finally, an acoustic model was proposed to integrate the velum effect for the speechsounds.