Hey all,

I’m happy to be making my first post to this community. I discovered it recently when embarking upon making a Magic Mirror and am impressed by the active support and discussion that occurs. Not sure if this post belongs in Development or Showcase as the module is mostly done, but I just had a daughter and may not be able to provide initial support for it so wanted to do a soft release here first.

I know there are existing voice modules for MM2 but I found my use case was to use Alexa alongside Sphinx, providing all the voice recognition functionality within one module. This prevents different modules competing for resources (microphone) and having to synchronize the allocation and deallocation of these resources. The hope is that non-audio related functionality will be handled by other modules, there is a not yet documented command with action “sendNotification” that will allow configuring certain voice commands to invoke a process in a different module.

Another goal for this project was to eliminate the need for lengthy configuration and setup. The module prefers convention to configuration, and most of the nerdy voice recognition and Alexa setup is bootstrapped for the user during the load. This allows the user to have a simple json configuration and the module will generate the appropriate ssh keys and language models/dictionaries required for the module to work.

Enough talk, here is the module. Pull requests warmly welcome, around documentation or functionality. Thanks.

https://github.com/whyjustin/magic-mirror-voice

So it its always listening in your version?

Yup, Sphinx is always listening (as long as Alexa is not listening). You can configure the command that invokes Alexa in the sphinx:commands dictionary. Any key in the object will be recognized by Sphinx and perform the action as configured by the value. For example:

"mirror mirror": {
  action: 'alexa'
}

Saying “mirror mirror” will trigger Alexa.

Note from admin: Please use Markdown on code snippets for easier reading!

Now the problem is that the module is not registering. I cannot connect to that link if I put it in a browser.

This is a long shot since you’ve setup Alexa already, but are you using a browser on the pi to connect to the site? If not you would have to replace localhost with the pi’s IP/Hostname

Also, I think it should be localhost:3000/provision/… because that’s how the alexa standalone app registration works but I could be wrong.

The oauth service can run on any port, I believe that 3000 may been taken by another process on my raspberry pi (either by Magic Mirror or something else) which is I why I used 5555. I think there is some redundancy in the three options (port/serviceUrl/redirectUrl) but it is required to keep the same port and hostname on all three. If you choose to keep 5555, you will have to update the Alexa settings that you previously set up to indicate the proper port:

https://github.com/alexa/alexa-avs-raspberry-pi/blob/master/assets/avs-web-settings-filled.png

@clumsy_ninja said in Yet Another Voice Module (Sphinx + Alexa):

@whyjustin Yeah thats it. Do i need to record a new training file to replace those or where did the original one come from?

The plugin will automatically generate your models and dictionary on startup using the JSON property names (e.g. “mirror mirror”) if the sphinx-model-generator/commands.dic and sphinx-model-generator/commands.lm are missing. So remove those files, adjust the JSON, then start the plugin again. You can also manually edit the commands.lm and commands.dic files after they are automatically generated if you are comfortable with language modeling.

@lb-rezende said in Yet Another Voice Module (Sphinx + Alexa):

Whoops! There was an uncaught exception…
TypeError: parser.close is not a function

Do you have any idea of what is happening?

Looking through the stack trace it looks like this is unrelated to my plugin.

@lb-rezende said in Yet Another Voice Module (Sphinx + Alexa):

Where do I set the $M2_HOME (and how)?

Looks like the $M2_HOME is set in the script

export M2_HOME=/opt/apache-maven-3.3.9

If the script resides in /etc/profile.d/ then it should run during boot (have you restarted your pi at all?).

Is there a way to uninstall those applications and do all over again?

Easiest would be to nuke the entire thing and start over. TBH these are linux questions and I’m very happy to support my plugin but they seem beyond the scope of that. I’ll try my best to help but may not be able to provide support on every question.

@lb-rezende said in Yet Another Voice Module (Sphinx + Alexa):

-Dmaven.multiModuleProjectDirectory system property is not set. Check $M2_HOME environment variable and mvn script match.

I would verify your Maven installation. Is $M2_HOME set? If so you can -Dmaven.multiModuleProjectDirectory=$M2_HOME although this should be the default behavior.

@clumsy_ninja said in Yet Another Voice Module (Sphinx + Alexa):

@whyjustin Yeah thats it. Do i need to record a new training file to replace those or where did the original one come from?

The plugin will automatically generate your models and dictionary on startup using the JSON property names (e.g. “mirror mirror”) if the sphinx-model-generator/commands.dic and sphinx-model-generator/commands.lm are missing. So remove those files, adjust the JSON, then start the plugin again. You can also manually edit the commands.lm and commands.dic files after they are automatically generated if you are comfortable with language modeling.

@lb-rezende said in Yet Another Voice Module (Sphinx + Alexa):

I’m not sure about the alpnVersion,

Verify your java version by running java -version then you should be able to look up the ALPN here. Your version looks incorrect so maybe this is your issue.

vlcPath and vlcPluginPath,

Check the docs here, but it should boil down to running whereis vlc. The values you provided look correct for a default setup.

because I do not know where and how to find it. The dsn line as well, where do I find it?

I don’t think you need to worry about this. It’s part of the automated cert generation and the communication between the OAuth NodeJS server and the Alexa Java application. IIRC this can be any value.

@lb-rezende said in Yet Another Voice Module (Sphinx + Alexa):

electron js/electron.js: …/src/java.cpp:105: v8::Localv8::Value Java::ensureJvm(): Assertion `result->IsNull()’ failed.

If I were to hazard a guess the JDK isn’t installed on your machine although you previously said that you were able to verify Alexa was working. Wild guess would be to verify the steps in [0] were followed to downgrade the electron client although this may not be related.

[0] https://github.com/whyjustin/magic-mirror-voice#installation-notes

@clumsy_ninja said in Yet Another Voice Module (Sphinx + Alexa):

@whyjustin Is it possible to change the activation work in your alexa? I remember seeing it defined in json somewhere but does it require voice training or something?

     commands: {
        "mirror mirror": {
          action: 'alexa'
        }
      },

If you edit the above you can adjust the activation of Alexa. Haven’t spent much time with this project recently but IIRC you have to remove sphinx-model-generator/commands.dic and sphinx-model-generator/commands.lm when changing the commands.

@clumsy_ninja said in Yet Another Voice Module (Sphinx + Alexa):

notice is has led’s indicating the status (listening, processing, speaking).
https://github.com/alexa-pi/AlexaPi

IIRC I exposed some of this stuff in the plugin, it shouldn’t be too hard to provide UI feedback for these events. Look at the events passed to the Java client here: https://github.com/whyjustin/magic-mirror-voice/blob/master/node_helper.js#L49. If you get anything working, PRs welcome.

What i really want to achieve is a module for the mirror which shows the visual feedback with a microphone icon and also show the GUI cards alexa service returns. Problem is I’m not sure how to make this hahah I’m only use to writing in C and Java.

GUI Cards are going to take more work.

@d3r said in Yet Another Voice Module (Sphinx + Alexa):

Your module is awesome! This is really good.

Thanks, good to hear it’s working. Let me know if you have any other difficulties or feature requests. As mentioned to @cowboysdude, it is easiest for me to track these through GitHub Issues. Cheers.

Now the problem is that the module is not registering. I cannot connect to that link if I put it in a browser.

This is a long shot since you’ve setup Alexa already, but are you using a browser on the pi to connect to the site? If not you would have to replace localhost with the pi’s IP/Hostname

Also, I think it should be localhost:3000/provision/… because that’s how the alexa standalone app registration works but I could be wrong.

The oauth service can run on any port, I believe that 3000 may been taken by another process on my raspberry pi (either by Magic Mirror or something else) which is I why I used 5555. I think there is some redundancy in the three options (port/serviceUrl/redirectUrl) but it is required to keep the same port and hostname on all three. If you choose to keep 5555, you will have to update the Alexa settings that you previously set up to indicate the proper port:

https://github.com/alexa/alexa-avs-raspberry-pi/blob/master/assets/avs-web-settings-filled.png