APPLICATION OF CONVOLUTIONAL NEURAL NETWORK IN AUTOMATIC SPEECH RECOGNITION FOR QURAN
Current automatic speech recognition system for Quranic utterance has low accuracy when it is facing testing data that is different with training data, either different speakers or verses. In previous researches, HMM-GMM is a topology that is widely used to build acoustic model. In this research, pe...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/27958 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | Current automatic speech recognition system for Quranic utterance has low accuracy when it is facing testing data that is different with training data, either different speakers or verses. In previous researches, HMM-GMM is a topology that is widely used to build acoustic model. In this research, performance of automatic recognition system for Quran is improved by using Convolutional Neural Network, a kind of deep neural network. <br />
<br />
<br />
<br />
<br />
<br />
Acoustic model that is developed using CNN with the same data to the Yusuf research (2016) can improve the accuracy. Improvement performance of HMM-CNN in closed scheme is 1.96% with 5.44% WER; improvement in different verses scheme is 0.58% with 9.17% WER; improvement in different speakers scheme is 4.24% with 9.76% WER; and improvement in different both verses and speakers is 5.07% with 16.93% WER. |
|---|