How we made Manuel, the tweet talking moose

1.

Time to come up with an idea for this year's Christmas card.

2.

Snowtime was many people's favourite Torchbox Christmas card (it allowed site visitors to activate a snow machine in our office). It was also one of the most fun for us, so we decided to build something else that connected online with our physical office.

3.

Inspired by the deer of Cornbury Park, the talking moose scene in Fawlty Towers, and Tom wanting to use 'Text to Speech', the idea of Manuel the tweet talking moose was born. So, we ordered a fake moose head from an ebay reseller in Canada, and called Tod (an engineer friend of Olly's, who can build anything from coil guns to bollock daggers) to talk about 'enhancements'.

4.

Matthew from our tech team hooked into the Twitter API to harvest tweets with #tbxmoose in, put together a node.js script to pipe them into the Festival open source speech synthesiser and got it all running on a Raspberry Pi, small enough to fit in a Moose head.

5.

Meanwhile Tod dusted off his animatronics chops, lopped Manuel's lower jaw off, added the necessary bits and bobs and put him back together again complete with fake teeth and lolling tongue.

6.

The audio feed attaches to this handy little Autotalk Interface Board, which moves the servo in response to audio or speech. The signal is split so it goes into the Autotalk board and into a speaker. Now we can hear Manuel speaking and see his jaw moving, at the same time.

7.

Back in the office Sarah and James designed and built the site, We rigged up a webcam, hooked it into Ustream (this is where the 10 second delay is between you tweeting and Manuel speaking, the rest of the system is almost instant) and got everything into place.

8.

A few, final tweaks - a profanity filter and some 'bon mots' to fill any long pauses between tweets - and Manuel, the Scottish, talking moose (who looks a bit like a donkey) was ready to receive Christmas messages.