I'm craving to know how the keyword spotter made by Google and implemented in several android-platforms ("ok google") is made, i.e. which methods (template-matching? deep learning? etc.) are used. Perhaps it not even available for the public :-) ?
Also, I was notified that Sphinx is working on a keyword spotter, is there a design for this yet? how long far are they in developing the application?
Best regards,
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm craving to know how the keyword spotter made by Google and implemented in several android-platforms ("ok google") is made, i.e. which methods (template-matching? deep learning? etc.) are used. Perhaps it not even available for the public :-) ?
I'm craving to know how the keyword spotter made by Google and implemented in several android-platforms ("ok google") is made, i.e. which methods (template-matching? deep learning? etc.) are used. Perhaps it not even available for the public :-) ?
Their system is described here:
Also, I was notified that Sphinx is working on a keyword spotter, is there a design for this yet? how long far are they in developing the application?
It is implemented long time ago. You can try it in pocketsphinx android demo. You can find description on on the wiki.
I'm more interested in the theoretical part/description of the system than trying the system. I've looked around on the wiki page for the description but fall short to find it. Can you perhaps show me a direct link to the description?
You seem very sharp in this field, therefore I want to ask you where I can read about (always-on) low power keyword spotting? I know there is a huge amount of litterature and implementations of keyword spotters, but I have searced desperately for those aimed at low-power devices.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello everyone,
I'm craving to know how the keyword spotter made by Google and implemented in several android-platforms ("ok google") is made, i.e. which methods (template-matching? deep learning? etc.) are used. Perhaps it not even available for the public :-) ?
Also, I was notified that Sphinx is working on a keyword spotter, is there a design for this yet? how long far are they in developing the application?
Best regards,
Their system is described here:
http://www.clsp.jhu.edu/~guoguo/papers/icassp2014_dnn_hotword.pdf
It is implemented long time ago. You can try it in pocketsphinx android demo. You can find description on on the wiki.
here is also a more recent paper:
http://www.isca-speech.org/archive/interspeech_2015/papers/i15_1478.pdf
On 11/13/15 10:52 AM, Nickolay V. Shmyrev wrote:
Thank you for your answer.
I'm more interested in the theoretical part/description of the system than trying the system. I've looked around on the wiki page for the description but fall short to find it. Can you perhaps show me a direct link to the description?
You seem very sharp in this field, therefore I want to ask you where I can read about (always-on) low power keyword spotting? I know there is a huge amount of litterature and implementations of keyword spotters, but I have searced desperately for those aimed at low-power devices.