German Voice Recognition

[maxbachmann]

Here is a first peak view at the voice recognition I currently work on for my MM project.
For the start I planned pretty much the same features like Mykle1’s Hello-Lucy
The recognition is in German, since I am German and we already have a ton of english ones
Like Mykle1’s it’s doing the recognition fully offline, but accepts full sentences and recognices your intent
Still in mid of work but wanted to share a first view on what I did so far (Forgive me the really really bad video quality)

https://youtu.be/jpM-4pq8dTs

Which commands beside the ones from Hello-Lucy would you like to see?

[Mykle1]

@maxbachmann

Nice work Max.

That is powered by a Pi? If so, decent response time.

[maxbachmann]

Yes runs on a raspberry pi 3b

[maxbachmann]

One thing I absolutely want to add is that my mirror is activated when I am in Front of it, but thats the speech recognition is automatically activated aswell so I don’t need to use the hotword while I am in front of the mirror

[Mykle1]

@maxbachmann

I thought I wanted my Hello Lucy to be listening for commands all the time, without having to issue the hotword, as you call it. What I found was that false positives became an issue, especially in a noisy environment. I hope you have better luck than I did with that. I am not unhappy having to say “Hello Lucy” and the false positives are all but gone.

[maxbachmann]

@mykle1 I would obviously need the hotword at least once to start the session and then listen for any intents for a while. Or in case of the mirror project just use the PIR Sensor as “Hotword” and then have the session going until the PIR does not recognise someone in Front of the Mirror anymore. Snips is actually a pretty good voice recognition. So far it even worked well with music in the background

[MrCoffee]

@maxbachmann
Great stuff!

Yes, there are only few german voice recognitions for MM2.
I really like your idea of only saying the wake word once or even get rid of it at all. I have got alexa running on the same pi3 as my magic mirror. It is not the Alexa-MMM-module. When I leave the mirror, the monitor turns off and mutes the mic. So alexa can´t hear a word anymore.
So if I am in front of the mirror, I could spend the alexa wakeword, I guess. Same with your wakeword.

Have you testet MMM-Voice-Control and MMM-Hello-Mirror? If I am not wrong, they can be used in german as well.
Not sure if they are on- or offline.

By the way: Do you have anything in common with Thomas Bachmann?
[http://blog.thomas-bachmann.com/2016/02/magic-mirror-2-0-mit-gestensteuerung.html](Link Adresse)

Cheers

[maxbachmann]

@MrCoffee No I have nothing in common with her, but saw the project a while ago

No did not test them actually. However snips is a lot more powerfull speech recognition going more towards alexa/google. MMM-Voice Control and MMM-Hello-Mirror are both more hotword detections. You give them a short sentence and they listen for it.

I made another short video of the current state: added the no hotword, a answer of snips (the voice can still be changed just the standard one any text-to-speech voice can be used and the always sam answer is definetly not the end aswell ^^)
The video shows aswell that you can really make normal sentences when talking to the speech recognition and it still reacts (in my opinion) pretty fast
https://www.youtube.com/watch?v=AhpehckdkjM&feature=youtu.be

[maxbachmann]

tomorrow I will add all the other Modules that are included in Hello-Lucy. Any other ideas of stuff on the mirror I could control by voice?

[Mykle1]

@maxbachmann

Other than hide and show? Most modules are not interactive. You could set up pages of modules. I’ve found that very useful.