A human machine speaker dependent speech interactive system

Speech has several characteristic features such as naturalness and efficient, spontaneous mode makes it an attractive interface element. It is possible to express emotions and attitudes. Its high bandwidth allows to develop multimodal interfaces. Telugu language has been used to communicate with the system. This process involves conversion of speech query into text, search for its suitable answer and then conversion of text into speech signal. So to perform the above tasks speech recognition system, search engine and speech synthesis system are needed respectively. Each of these modules takes input and passes output to another module in sequential order. First the features of the raw signal that are stored in the speech database are extracted and given to the speech recognizer which produces its equivalent text form and sends it to the search engine, whose duty is to find suitable answer from the database. It passes answer to the speech synthesizer that converts text into speech (i.e., reverse operation of speech recognizer) and sends back to the user. Speaker dependent speech recognition systems give more recognition performance than speaker dependent systems. In order to build a speaker dependent speech recognition system the data was recorded by a single male speaker and also for speech synthesis 1 hour speech is used. The data was recorded in a computer laboratory and background noise was normal. This system works using only microphone and our next step is to build a telephonic communication system.