Isolated Speech Recognition over Internet Protocol in Distributed Systems

In this study, isolated speech recognition system on client/server structure is implemented. In this system, client and server communicate through Internet protocol (IP). The client performs speech analysis and sends the speech feature vectors as a payload in UDP (user datagram protocol) packets to the server. The server stores incoming vectors and runs speech recognition algorithm when the last packet of speech is received. The server also performs packet loss concealment (PLC). The UDP is an unreliable protocol and the UDP packets might be lost during transmission. Therefore the PLC is used to improve the recognition rates. HMM (Hidden Markov Model) is used as recognition algorithm in the system. The system is evaluated with ESOGU isolated word database. In the experimental study, it is seen that the replacement-based PLC technique improves the recognition rates.