Voice Authentication (using Hidden Markov Model) and Siri/alexa like voice command responder (using tf dataset)
Amrita School of Engineering, Bangalore, has this hackathon every year, called SLAC. This year about 250 teams had submitted their ideas and about 80 teams were qualified for the main hack.The hackathon was held in Amrita for a duration of 24 hours, on 30th and 31st of march. The 1st place cash prize was 15,000 rupees,and 2nd place was 10,000 rupees.
Our team comprised of Nishanth D A, Mahadev M and me(Vijay Koundinya). GE Healthcare , the main sponsors of the hack, had given 4 problem statements to choose from.We chose one of them and submitted our abstract idea. It was mainly about user voice authentication + Google voice assistant like voice command responder using speech recognition. We built an Offline voice command responder using the tensorflow dataset and made it run on an android app.For authentication we used a HMM(hidden markov model) approach, which could also run offline.We also a built a flask application to run the model on a computer. Judges were very impressed with our approach and implementation of the idea to the problem and the future extensions of the project.They liked the scalability and the multi platform approach we had taken(Web and Android Application). We secured first place in this hackathon, winning 15,000 rupees.
Problem Statement we chose:
Compatible System Assistant: Similar to Alexa, Siri, develop a voice control package which could run on System without connecting to Internet with a very limited voice command set to authenticate user and authorize him/her to control the System. Also, this package should be customizable for each product which should accept its command set as an input and also should be fault-proof as it frustrates the user to repeat the commands when the product doesn’t recognize them.
Background: The medical systems offered by GE are mostly used in sterile/wet environment and often user needs to interact with the system by using the touchpad/keyboard which is difficult as hands could be wet/need to re-sterile after using the system, he/she could be wearing gloves and the touchpad doesn’t recognize the input of the user.
