Alexa on Rasberry PI- DIY Amazon Echo

[johngalt]

This looks pretty interesting. Not as easy as buying an Amazon Dot, but cheaper.

Apparently this might be old news, but it was news to me.

http://www.bbc.com/news/technology-35899887

 

[pete_c]

Here utilize an Alexa (on demand) and Kinect plugin on one Windows 10 machine that serves as a large touchscreen interface and works fine.
 
Using the Alexa plugin allows me to connect to my zoned audio stuff.  This is related to using Homeseer 3.

 

[wuench]

I saw this and was about to pull the trigger using an old Pi but it doesn't support an always on microphone.  Which was a deal breaker.

 

[BobS0327]

I saw this and was about to pull the trigger using an old Pi but it doesn't support an always on microphone.  Which was a deal breaker.

Unfortunately, Echo only implements on-device keyword spotting   to detect the "wake word".  When the device detects the "wake word", it streams audio to the cloud.
 
Thus, the "wake words" such as Alexa, Amazon are inherent to the Echo device.  The Amazon Voice Service (AVS) has no "wake word" implementation.  Same thing for the Amazon Skills Kit (ASK).  So, the only options are to implement a physical push button as indicated in the RPI project or integrate a third party speech recognition package such as Sphinx  http://cmusphinx.sourceforge.net/  with the RPI project.  The 3rd party package would "listen" for the "wake word" which would in turn activate the AVS "listening service" upon "hearing" the "wake word".
 
Using the latter option, gives you more flexibility in implementing a customized "wake word".  For example, you can say "PI, turn on kitchen lights".

 

[johngalt]

Unfortunately, Echo only implements on-device keyword spotting   to detect the "wake word".  When the device detects the "wake word", it streams audio to the cloud.
 
Thus, the "wake words" such as Alexa, Amazon are inherent to the Echo device.  The Amazon Voice Service (AVS) has no "wake word" implementation.  Same thing for the Amazon Skills Kit (ASK).  So, the only options are to implement a physical push button as indicated in the RPI project or integrate a third party speech recognition package such as Sphinx  http://cmusphinx.sourceforge.net/  with the RPI project.  The 3rd party package would "listen" for the "wake word" which would in turn activate the AVS "listening service" upon "hearing" the "wake word".
 
Using the latter option, gives you more flexibility in implementing a customized "wake word".  For example, you can say "PI, turn on kitchen lights".

 
That is really interesting.  And for those that are concerned with the privacy of a Echo always listening, if Sphinx doesn't rely on the cloud then you don't have to worry about Amazon "always" listening.
 
Me personally will likely just by a Echo, plug it in and call it a day.

 

[batwater]

If you have Android devices, AutoVoice for Tasker allows for the definition of any wake word or word phrase to trigger processing

 

[BobS0327]

if Sphinx doesn't rely on the cloud then you don't have to worry about Amazon "always" listening.

 
Sphinx is a local app.  It doesn't rely on the cloud for any functionality such as a recognition engine etc.  From the HA perspective, using Sphinx has a downside in that the command set has to be very minimal.  Using an extensive command set would, IMHO bog down any "beefy" desktop system or back end server.  But I do plan to use Sphinx in my current ongoing smart home project subject to this limitation.
 
I think a cloud based speech recognition system would be more applicable to the more advanced platforms such as the Amazon Echo.  For example, I would never attempt to  use Sphinx to update and analyze  my grocery shopping list, especially if I was using IFTTT with my shopping list.  I don't know if that is even possible with Sphinx.  So, Sphinx will be used for the simple HA commands such as turn light on, turn thermostat  up to 72 etc. Everything else will be handled by Alexa.