This is a bit of a fun project I’ve done in recent weeks. Actually adding machine learning for the AIY Voice kit. Most of the ideas I got from Python Programming with Deep Learning.
What I did was install tensorflow on the ladybot and set up a deep learning neural network which is able to gauge the sentiment of the sentences sent to the the AIY Voice Kit. This is based on a huge dataset of positive and negative statements gathered from tweets.
The building of the neural network takes a huge amount of time – about a week 24 hours a day. But luckily this is also available for down load. The best way I found of implementing the system was by setting up a socket to the neural network and then feeding the text from the ladybot into the socket. Other methods didn’t seem to work. Code can be found here aiy with Machine Learning.
Suprisingly this works really well with high accuracy rates and is very fast in returning a response. Not sure how this could be used practically but shows that there is potential for this type of system. Here is a link to a demo on you tube AIY ladybot with Neural Network. All the sentences spoken are unplanned.