An in-car oriented service that plays audio news personalized on passengers' interests.
Consider the situation in which you are traveling with other people that have different interests, and you do not know what to talk about.
You could play some radio news or listen to pop music, but maybe you and the others share some common interests to specific topics. It would be great to discover it and talk about them!
TravelTales is a service whose aim is to stimulate conversations between people in a shared space.
The service was developed for in-car usage.
The main characteristic of TravelTales is to know people preferences and propose news that are the nearest to the interests of the passengers.
The service can fetch news from any website, then for each of them it generates a summary which will be indexed according to topics that are treated in that news.
For each news summary the system exploits a TTS model to produce a synthetized audio version of it.
Every time a TravelTales client instance tries to load the most interesting news for the group that is recognised as currently onboard, the retrieval system loads the interests of the passengers and computes the ranking of the most interesting news.
At that point news audio file is played by the client system.
After news proposition, the system starts a feedback gathering phase in which people engagement level is estimated by exploiting audio and video features coming from each user.
The system is capable of recognizing who is speaking, the vocal emotion and also the facial expression of each passenger onboard during the gathering phase.
Some short video demos:
The code is freely available on GitHub, in this repo.
In order to run it, simply follow the instructions that you can find inside the repository.
If you encounter any issue or problem during the deployment process, please contact me, I will be glad to give you more detailed instructions.