BROWSE
Detail
Conference Paper
Performance Improvement of the Conversational Speech Recognition System using Deep Neural Network in a Car Navigation System
Cited - time in 
Share 
Share
- Authors
- Woo Yong Choi, Kiyoung Park, Hyung-Bae Jeon, Jeon Gue Park
- Issue Date
- 2016-11
- Citation
- International Conference on Engineering, Technology, and Applied Science (ICETA) 2016 (Fall), pp.1-7
- Language
- English
- Type
- Conference Paper
- Abstract
- In this paper, we construct the commercialized speech recognition car navigation system in the real world. We configure the server/client model for conversational speech recognition. The Acoustic Model (AM) uses Gaussian Mixture Model (GMM) to represent the probabilities of the Hidden Markov Model (HMM) states. To find out the effect of the use of real-world data in training model, we retrain the GMM-HMM with the log data. The accuracy of the speech recognition server is improved. We obtain 44.7% Error Reduction Rate (ERR) by updating the GMM-HMM. Recently, Deep Neural Network (DNN) is spotlighted in the speech recognition field. Hence, we construct the DNN-based speech recognition server, which replaces GMM with DNN.The experimental results show that the DNN-based system outperforms the other GMM-based systems. We obtain additional 46.9% ERR compared to the updated GMM-based system.
- KSP Keywords
- Car navigation system, Conversational speech recognition, Deep neural network(DNN), Error reduction, GMM-HMM, Gaussian Mixture Models(GMM), Gaussian mixture(GM), Hidden markov model(HMM), Log data, Real-world data, Speech recognition system