An Application of Auditory Periphery Model in Speech Processing
Biometric identification system using speech signal is a demanding research field in signal processing because of its growing applications in e-commerce, technology, and crime investigation.
The motivation of this study is the human auditory system accuracy in audio signal perception and identification. The availability of digitally implemented Cascade of Asymmetric Resonators with Fast-Acting Compression (CARFAC) cochlear model and the deficiency in automatic identification system performance of the existing methods in comparison to the human performance inspire me to explore the CARFAC model in automatic identification tasks.
This study presents the application of the CARFAC model in automatic speaker identification, gender detection and accent identification system. The real-time hardware implementation of the automatic speaker identification system is another goal of this study. Gaussian Mixture Model with Universal Background Model (GMM-UBM), Support Vector Machine (SVM) and Identity-vector-Probabilistic Linear Discriminant Array (i-vector-PLDA) will be used to assess accuracy of the propose system.
Three different data-sets will be used as input speech signal: Bangla, UM and Australian accent data-set. The AusTalk data-set will be used only in Australian accent identification task. Auditory Nerve (AN) model proposed by Zilany (2006), Mel-frequency Cepstral Coefficient (MFCC) and Frequency Domain Linear Prediction (FDLP) will be used to present baseline methods in automatic identification system.
- M.Sc in Biomedical Engineering
- PhD Candidate
Contact Md Atiqul