Prototyping weeknotes #87
It's my first time writing weeknotes and there's lots of interesting progress to report from around the team.
Barbara reports that she had a very good week in FI-CONTENT-land with UX proceeding to scope out work. She then dashed over to Brussels with George for a formal project review at the European Commission. Happily, they got very good marks.
Chris and Yves worked on the Roar to Explore classification prototype, which takes the sounds of approximately 1000 examples of children making animal sounds and attempts to extract "features" from the audio. Using 21 MFCC coefficients together with a single gaussian covariance matrix gave an accuracy of around 75% and led to a week of guesswork in the office between owl and elephant noises. Chris is hoping to talk to anyone who has experience of selecting audio features for this kind of task, so leave a comment or get in touch if you do!
Yves also worked on speech/music segmentation, writing a Vamp plugin for it. Vamp is an audio analysis plugin format and his work is on Github. Using the plugin, Yves was trying to derive a measure of how much music there was in an individual programme, to get a rough idea of whether we would have the rights to put them online, so it's not very accurate. It's quite fast though, as we intend to run it on a very large collection of programmes.
Recommender algorithms abound with Chris Newell completing a comparative study of recommender algorithms based on minimal user feedback. Collaborative filtering appeared to give the best overall results.
Meanwhile, Becky launched the first part of her internal Online Recommender Field Trial, which is to gather user feedback to help evaluate the performance of a number of recommendation algorithms. So far the trial is running fairly smoothly, and she is just starting to do some preliminary calculations with the data, which are showing some interesting results!
Theo presented the "News Follow" project findings to the News User Experience and Design team, which was well received. He's also been working on LIMO, writing up UX for the documentation and preparing a proposal for where to take the project next. I've been taking Theo's designs for the LIMO demonstrator and implementing them using a heavy dollop of CSS3. Chris Needham has been working with Sean on the GStreamer live stream system that will be hooked up to the demonstrator interface. Chris made a breakthrough late on Friday afternoon, leading to much wooping in our IRC channel.
Vicky B. has been in a cycle of planning/refining/practising for user testing of NoTube's N-Screen prototype. Vicky and Libby will be testing this second screen application for choosing what to watch with 10 participants in the Centre House user testing lab next week. They hope to find out how people respond to the different types of programme suggestions in N-Screen, and to learn more about how sharing suggestions with other people in a small group using N-Screen can help the group decide what to watch on TV together.
After kicking off the week with a visit to BBC Monitoring, Tristan chatted with the Future Media Radio & Music team about their plans and priorities and is going through some final checks before we can make the Programme List alerts live, which is the most-requested feature. He's also been wondering if there's a better alternative to the 5-star rating system?