Gender identification of children using hidden Markov model based on Mel-frequency cepstral coefficient / Adira Ibrahim
Speech is a communication between humans using variety of language that is translated into word, phrases and sentences. Speech signal carries pitch intonation that can express information such as accent, emotion, gender, and age. However, study in vowel for children has some difficulties such as fal...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2013
|
| Subjects: | |
| Online Access: | http://studentsrepo.um.edu.my/7825/4/adiraibrahim_KGL110004.pdf http://studentsrepo.um.edu.my/7825/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Speech is a communication between humans using variety of language that is translated into word, phrases and sentences. Speech signal carries pitch intonation that can express information such as accent, emotion, gender, and age. However, study in vowel for children has some difficulties such as false pronunciation and disfluencies of speech. This project aims to develop a system that can identify gender of speakers based on speech signal using Hidden Markov Model (HMM) as a recognizer. Mel Frequency Cepstral Coefficient (MFCC) was applied as the feature extraction method. HMM was trained with Baum-Welch algorithm and tested with Viterbi algorithm to get the gender identification accuracy. For single frame analysis, maximum accuracy was obtained at 64.17% at signal length of 30ms. For multiple frame analysis, maximum accuracy was achieved at 64.26% at AFL 20ms with 10 ms shift. For the single frame analysis, the accuracy of female children was 67.78% while accuracy for male children was 60.56%. For the multiple frame analysis, the accuracy for female children was 65.74% and 62.78% of male children. Hence, female speakers had higher identification accuracy compare to male speakers. |
|---|