Question about offline nature of Snips

offline

#1

Hi guys

Just found Snips after doing some research on offline voice assistants.
Sounds very promising and I might be interested in getting the maker kit and playing around with it.

I am a little confused about the offline nature of Snips - is it really offline?

Reading articles like this:


Which says:

Developers can offload individual elements such as wake word processing or natural language recognition…

and then posts like this:


…makes it sound like its kinda online - kinda offline?

For Text To Speech:
Is that offline? (I don’t mind robot sounding voice… doesn’t have to sound like an Australian Alexa)

For Speech To Text (voice recognition):
Is that offline? Or if there is an offline mode / option for this - is it any good? (I’ve read bad things about Sphinx, etc…) Does it use the NLP? or is the offline mode some lesser option?

Skills:
From what I understand, you configure skills in some online profile, and then download the skills to your Snips device - so that sounds offline (I understand that the individual skills might need to access the internet to to check the weather, access Spotify APIs, etc…)
But - its not sending voice comments back to Snips HQ where it matches the command to the skill, right?

Thanks!


#2

It’s entirely offline, completely on device. Turn off your wifi and it works. But how do you want to grab the weather forecast when being offline? So you can use online apis


#3

As Psycho Said, it’s really 100% Offline, all components work offline as default.
The good thing is that the modularity of Snips allows some parts - if you choose to - to be “improved” by using online APIs/components, meaning if the voice sounds too robotic to you (I don’t - FR voice), you can use another one (cloud one…), etc… You can do whatever you want with it : “the sky is the limit” :slight_smile:
Of course, some specific data are requested online when needed (weather, etc…), if you choose to install those specific App that needs Online data, but only pieces of data : the code/App runs on your device. No online data needed : 100% offline.

I personally prefer to be 99% offline (see my post on SnipsFreeWord), and it works perfectly as it is, and by that, I don’t mean it’s “OK it does mostly what you want”, it’s a real alternative to Alexa/Google personnal data siphon solutions, providing you have some understanding in computing for the time being, as it is yet a young product.


#4

Thanks guys

That is reassuring.

I found a lot of contradictory info during my 1 day research excursion into Snips-

https://medium.com/@zibra_/how-to-build-a-voice-controlled-speaker-that-protects-your-privacy-ec6429a2c673

Configuring the Speech-to-Text: To enable speech to text using Google Speech to Text…

Everything runs directly on-device, meaning no one will ever hear your voice


Thats kinda confusion…

Thank you anyway - this definitely seems like a worth while project to try out. I’ve already ordered a cheap USB mic for my pi to test it with. Once things get going, I’ll upgrade to a better mic.


#5

It’s not really confusing. You have the possibility to enable google speach component if you wish so


#6

It’s not confusing when you know it.
English is not confusing for someone who speaks it. Try learning another language, then see if it’s confusing or not.


#7

Dude, I speak french, english, swiss german, german, I learned russian, I speak finnish with my wife and I understand pretty much of italian… When I say it’s not really ocnfusing, it’s about the fact that the components aren’t. I’m not talking about the languages specialties


#8

It was an analogy. No need to get all upset.