Â鶹ԼÅÄ

Devices

The devices team updated for this year’s Glastonbury festival. With the social media promotions reaching 81% mobile users, Andrew has updated the app to work on Android and iPad devices. Overall, the app has received a 4.32/5 rating, with the Taster page receiving 5.5k hits. 

Further investigation was done on the Eurovision prototype for the project. Andrew has produced architecture diagrams for each of the three prototype ideas (EuroTime, EuroQuizion, and EuroMeme). Chris has been looking at which of the partner technologies, together with our own work on device discovery and authentication, could be used to build the prototype.

Chris and Joanne visited MCUK to observe the user testing of the current iteration of user sign-in on TV iPlayer, using the (CPA) pin and pair flow. Overall, users found the process to be confusing, but there are user interface changes that could be made to improve the process.

Chris is continuing to be involved in the . The group have recently been looking at integration with HTML5 video, conditional access and DRM systems, and programme recording.

Andrew and Chris set up a server on the internal R&D cloud to build the custom Yocto Linux distribution for Radiodan. Thanks to Tomas Frydrych's good work, Yocto was easy to install and set up.

Andrew has been continuing development of a small Radiodan application that allows capacitive touch and RFID sensors to control media. The app uses the Espruino electronics platform and React as the web UI. It will be used in a collaborative workshop with UCLIC to observe how end-users creatively customise media applications.

Libby, Andrew and Dan have started planning to demo at in September.

Joanne, along with the rest of the UX team, received training in Domain Driven Design from .

Data

In the data team, James has continued research into  (LSH) a technique that Yves used for the . The hope being this technique can be used to speed up queries within the CODAM search engine.

Matt was able to train and test his own model for the Kaldi software, this starts with some audio and transcripts which are used to build a complete system for speech recognition. It's not the most advanced or accurate of systems but does already offer significant improvement over previous work with . 

Jana has spent some time reading up on the methods used in Kaldi and hosted a tutorial for Matt, James and Chrissy. It ended up being a whirlwind tour of the whole field and gave the team loads of starting points for further research.

Ben has been doing a literature review of methods for automatic case restoration and punctuation, it is hoped this will give the ability to restore the text from our STT systems.

The data team focused on sharing knowledge this sprint, by setting up a  group for sharing papers and research. Ben has also been using the mind mapping software , which can be used to easily create diagrams and share them via Google docs. Jana hosted a talk on CODAM for the 1ES / UCL Show and Tell. It was well received with attendees from other R&D sections and UCL departments.

Content

For the content team the next iteration of atomised news research began, by taking forward the card-based prototype. Collaborating with , research has began into finding production tools available and what could be developed that would help a journalist push content onto the card app. The content team have already defined the stories to work on. They are now deciding on a data structure and are finding a journalist to work with on creating the content.  

Discovery

Work on Mango, our entity extraction tool continued, Thomas and Chris along with Francesco from Â鶹ԼÅÄ News Labs have been working on Docker images and automating Debian packages after a successful build. They also looked into defining a strategy for multilingual model updates. The overall goal is to make it easier for other teams around the Â鶹ԼÅÄ to deploy their own instance. They have also worked with Dong Liu from Â鶹ԼÅÄ K&L, who has been looking at the updates Mango needs to support other languages that use different alphabets to English.

The discovery team published a maintenance release of the audio visualisation library . Thanks to generous external contributions, version 0.4.5 brings some bugfixes due to the Kinetic library. Peaks.js now uses Konva, a maintained fork of Kinetic, which has been deprecated. This version also introduced configurable zoom adapters and more configuration options for axis label and gridline. (More detailed informations to be found in the of the project.) Its pleasing to see Peaks.js is used by the Rewind Radio project as an enabler to share radio clips with your friends.

The discovery team have been working with (AWS). Devising the simplest way to deploy prototype sites to AWS behind an SSL certificate, using their Heroku-like (PaaS) solution ElasticBeanstalk.

Interesting links