• Extracted both time-domain and frequency-domain acoustic features (e.g. MFCC, LPCC, GFCC).

• Implemented adaptive segmentation within front-end pre-processing using multiple features.

• Developed Deep Neural Network - Hidden Markov Model mixture model, the classification accuracy improved by 8\%.

Supported by IBM-research China

• Coordinated with three partners, brought a new product from concept to entity.

• Established a speech recognition module with background and GUI version, using iFLYTEK.

• Set up local server on host computer to enable cross-platform communication.

• Based on TCP/IP, utilized Socket to communicate with other parts, like PLC and video processing module.