-Generally in meetings/lectures you cannot just focus on meeting and give your full attention to it as you have to take notes or else you will forget key points of the meetings. Also the sound recorder in our mobile app are also not smart enough to let us search our recordings based on text of recording!
What it does
-ATRIS is driven by the goal of adding value to human conversation. People discuss numerous things in a discussion, it may involve task allotment, mentions of key entities, dates and various other components. It is demanding to keep track of all the details over the discussion, in scenarios of discussion between doctors-patients, corporate meetings, missing over details may result in a predicament. Here ATRIS comes into play we bridge the gap between audio and text representational form. ATRIS effectively transcribe the audio to text in real-timex and perform various Natural Language Processing on the result to generate an effective report in an interactive dashboard where people can see the audio transcribed data and analysis with all key entities discussed over the discussion.ATRIS is the missing part of AGILE or any sort of discussion and task allotment process, as it results in increased overall productivity of the entire discussion. We are able to keep a record of the discussion which can be referred over the period of time to know of any discussion which was processed by ATRIS.
How I built it
We have built a web app which is built on react as frontend and the backend is built on python Django and all the machine learning implementation is done through pytorch and applications are like text summarizer,entity recognization, keyword recognization and sentiment recognization and as well as sound classifier and also used model implemented in pytorch for speech to text
Challenges I ran into
Many challenges like socket connection from backend to frontend, implementation and making of pytorch, managing meetings database and scheduling all tasks in a queue and their chronology and many more.
Accomplishments that I'm proud of
I m proud of that even though there were many obstacles we were able to complete the project and were able to use pytorch in our project and have created a working prototype.
What I learned
I learned to make models in PyTorch , sockets in django, celery and doing a great team coordination to carry the work.
What's next for ATRIS(Automatic Transcribe Intelligent and Smart)
The plans are very exciting, ATRIS will be developed into mainly two types of products: Cloud based b2b general product or a custom on premise solution. The later would be much suited in the corporate setup, as the cloud based product will be able to serve the corporate needs like remote collaboration on the discussion and group editing and sharing features. The later product of custom on premise solution would be much suited for places where a lot of specialized domain knowledge is there in discussions, like medical documentation conversation between the doctors and the patients. ATRIS would be able to impact the education domain to a great extent, students would not need to worry about documenting the general dictation of lectures and can rather focus on understanding and organizing their knowledge base according to their needs.