Speech based power outlet controller

The goal of this project is to develop a completely offline based limited domain automatic speech recognition system that can recognize <100 commands with >90 percent accuracy in a generic noise environment.

This system is designed to be offline to avoid service disruption due to non availability of network, privacy concerns, Independent solution. The system is made language independent to make it available readily for users in other languages for which huge databases are not readily available.

I decided to make it because there are not many products available in the market to perform Offline ASR even though powerful computing architectures are available because there is no API for doing so (this project work is part of my research work).

The system takes input from a 2d spatial beam former microphone array whose input is processed through a delay and sum beam former and a couple of speech enhancement algorithms then the data is passed through a custom ASR algorithm currently under development in our lab.

The command output is then given to a power outlet controller that performs the required power switching tasks.

This system can be made in to prompt less ASR system but will work better on a prompt initiated recognition.

Status message sound synthesis is also made available to advise acknowledgements or give custom status messages.