Of course, the Arduino computing power is too small to run a real speech recognizer. So I am running a speech recognizer on a 1GHz ARM-based test board with A13 chip. The board is behaving as an Arduino shield. It communicates to the Arduino Uno board via a serial link. In the first version, I trained the recognizer to control all the pins of the board (analog and digital). The default can be extended by adding your own trained sentences.
I then quickly had the idea of making these sentences changeable. The first prototype came to live.
To make it mechanically more stable, I integrated everything onto a several prototype boards.
I presented that together with some friends and business partners at Maker Faire 2015 and received awesome feedback from literally hundreds of interested hackers.
However, making it one board, was still a challenge. In order to that I needed professional help. So I got together with Bertrand and run a Kickstarter campaign. Amazingly enough, we got 233% funded!
With the money, we were able to make it all into one board. Here is the Eagle CAD of it:
Having done this, I worked on creating an open-source Arduino library (see Github link), which hopefully makes it easy for everybody to train their own sentences and dialogs. This library comes with a couple examples, including a light switch, an Eliza bot, an alarm password system, and a voice controlled computer game.
Then it was: Testing, testing, testing! So many Arduino boards and variants!
We then sent out the boards to our about 350 Kickstarter backers.
What now? Well, please go ahead and try one yourself! Also, I am going to publish about projects with it and about it on my Hackaday feed!