Inspiration
Ever been on a date and you are trying to figure out if your partner is actually interested in you or if they are just pretending? It is hard to decode the subtle hints in the language and sometimes we may interpret a casual "yes" for positive and the date may go horribly wrong. To solve this, we use machine intelligence to decode the subtle feelings in the conversation and give the user a feedback, in real time as the conversation happens. So all the best with acing that date.
What it does
- The application records speech and converts it to text using google speech-to-text engine.
- The text is then parsed and fed as input to a Watson engine running sentiment analysis service.
- Watson engine decodes the subtle emotions in the conversation and provides scores for various sentiments.
- We use this to inform the user on how the date is going and provide feedback to make it better.
Challenges we ran into
Integrating various systems and making them work together that too in real time was an issue. Had to eliminate the bottlenecks through out the process so that the end-to-end system runs as quick as possible.
Accomplishments that we're proud of
Proud of making an utility which we can use as soon as the next weekend (or whenever we get a date :-)) and use it to improve an aspect of the effectiveness of a date which we feel is a whole lot in grey area still. Also, getting a system to work as a whole by understanding various dependices was real fun.
What we learned
Things which seem super simple may not be so simple until you actually get it working. There were times when we were stuck debugging a supposedly "silly feature" and spending hours on it to get it working, which we had thought earlier would be a piece of cake.
What's next for Date-o-meter
We could not find enough time to find an effective and non-intrusive way to distinguish between the conversation of the user and the partner. Ideally the application should listen to only the partner's conversation and decode it but right now it records the conversation of both parties and scores it. We can get around this if the user starts and stops the recording only when the partner speaks, but that's annoying and by our guess is your partner would run away by looking at you record their conversation. That's the last thing we want our app to do. We will try to figure this out in future but for now the app decodes the whole conversation.
Log in or sign up for Devpost to join the conversation.