AN ENTROPY CODING METHOD OF CONDENSED FEATURE PARAMETERS FOR DISTRIBUTED VOICE RECOGNITION IN MARINE ENVIRONMENTS

摘要

PURPOSE: An entropy coding method of a compressed specific parameter for distribution voice recognition in marine environment is provided to add and transmit 10.07 bits as an error prevention bit, thereby preventing performance degradation of a distribution voice recognition system caused by packet loss during communication of the marine environment. CONSTITUTION: A distribution voice recognition system quantizes an MFCC(Mel Frequency Cepstral Coefficient) and log energy extracted in a corresponding frame. The distribution voice recognition system additionally compresses a sub-vector index of the MFCC and indexes of C0 and log-E sub-vector by using a Huffman table corresponding to voice classification of the frame. The system calculates differential information for the compressed index. The system applies differential Huffman coding to the differential information.