On device computing
Small memory footprint and cross-platform support
I have always used pocketsphinx for Hotword recognition in my app. The problem was it was too heavy on resources- especially battery. https://snowboy.kitt.ai/ is a great alternative.
So, you can use this lib to detect hot words. Well, then what?
It's great that it can detect hotwords, but after the hot word, the user is then going to speak a sentence/command for my app/gadget to do, and won't I then need a full-fledge speech recognition engine to know what the user wants my gadget/app to do?
And if I do need a full SR, is there a speech recognition engine for RaspPi that works offline?
So, you can use this lib to detect hot words. Well, then what?
It's great that it can detect hotwords, but after the hot word, the user is then going to speak a sentence/command for my app/gadget to do, and won't I then need a full-fledge speech recognition engine to know what the user wants my gadget/app to do?
And if I do need a full SR, is there a speech recognition engine for RaspPi that works offline?
I was wondering about that too. Without the SR software being able to recognize common words of a language like English, it will be limited to predefined words to access your views such as buttons. It would work best when no data entry is required, or the data entry is just numbers. It does have a use, but it won't do dictation. If it did do dictation, the data bloat would be enormous.
Yes, you have to use snowboy with speech recognizer.
please have a look at the the video in the First post if you have not already. The goal here is to keep the microphones active, so as to recognize just a particular set of words. This is different from how a speech recognition or pocketsphinx works.