Just to clarify with some context:
This voice based keyword detection requires an mbed discovery board; and I am sure some min flash size.
Even for a hobby project, i.e. personal home automation, putting several discovery boards in various rooms is costly and not ideal.
When answering the question, please consider the following two use cases:
1) What do business do for going from PoC to production?
2) What would hobbyists do that is similar to a business but on a smaller scale of course?