Murmured Speech Recognition Using Hidden Markov Model

When criminals or militants are injected with a dose to extract the truth, they murmur the truth. It is very difficult to understand by human ear. Also in war field when the commando has to give confidential instructions to his soldiers present at distant locations. This paper proposes a method to capture, send and convert Non audible murmur (NAM) speech to ordinary speech. To capture speech data from human beings behind the ear as whispered voice or non-audible murmur, NAM microphone is used in this paper. This murmured speech can then be transferred via wi-fi transmitters for voice Conversion and detection systems. A quality articulated murmur is captured by the NAM microphone progressively associated with wi-fi handset is attached backside of the ear of a murmured human. This kind of set up can also be used for communicating securely. The output speech is robust against the environmental noises, because NAM microphone sequentially connected with Wi-Fi hand set directly transmits the signal to conversion and recognition system. Speech recognition system in this work uses dictionary of the murmured voices to achieve better accuracy in recognition. Generally, in the mechanism of body conduction, a poor quality of voice is achieved so far. This paper proposes a trained state transition conversion model to improve the quality of speech based on Hidden Markov Model (HMM) through body conduction.