An Adruino might be an option... given it does not use a complete OS, but just C source code for the most part. There are a lot of real-time solutions that use Adruinos.
When getting down to C it might be time to look at a DSP. I know that Alexa uses a DSP for the wake word, then streams the remainder of the sound to the cloud. I would expect Google and others to do something similar.
1 Like
It is a matter of complexity, right? If you don't want to leverage the cloud? And just do something simple, keeping it entirely local might be an important consideration. IMHO, if you can avoid engaging the cloud, you should.
1 Like
Hi all!
Just found out that Picovoice now has a free tier! https://picovoice.ai/
Will give them another try!
2 Likes
Then you are one of the (few) lucky ones, when looking at the very limited set of supported languages 
1 Like
Let us know how it goes. I find their docs confusing. One place it says free wake words but in the console it says 30 days... Not sure which is right.
Thanks
What I'm missing here is the use case. There are many options but to pick the best solution I think you need a use case. Nevertheless, just to add an option you haven't listed:
- Create an Android app and use it's built in speech to text capability as that one is for free then -though it's not as good as their shiny new cloud service.
- Send the text via http request to your server running your nodered flow.
However, I'd also send it through an nlp tool so that you get structured data, extracted from the natural language text. That may happen before sending the request to the nodered flow or the nodered flow itself may call an nlp tool to do the job.
There are many big nlp tools out there, like nltk, opennlp, corenlp and even more hobby ones like mine 
Just noticed that I replied to a very old thread which already has many good replies but somehow popped up in the recent list...
1 Like
Hi there - Sorry for a very late response, I think you have already figure it out, but here's the answer: you can train free wake words but within 30-day window you can train max 3.
Although we want to help everyone, we're a startup with limited resources, we are only able to offer support on Github. If you have any feedback on docs, you'll see the GitHub icon on every page (On-device Voice Recognition Intro - Picovoice Docs) please create an issue and help us fix or clarify whatever is not working (in general you can leverage GitHub community support for any technical question if docs not helpful)
and yes I'm working at Picovoice. 
Unfortunately, I just saw this page and wanted to close the loop in case it helps future readers.
3 Likes
