ROS Resources: Documentation | Support | Discussion Forum | Service Status | Q&A answers.ros.org

Voice recognition without a cloud.

Are there any packages or successful implementations of a stand-alone system for processing simple voice commands. While systems using the Google voice API and AWS clouds are widely used, networks are not always available for voice processing. Pocketsphinx worked back with Indigo, but I need something that works with Kinetic or higher.

I have been unable to install pocketsphinx for 16.04. and systems like Simon require training models.

Thanks.

2 Likes

used to work on in-cockpit conversational voice applications for automakers, not all of that is applicable, not all of those solutions are available to you. Besides pocketsphinx these might be worth a look …
https://www.soundhound.com/houndify


I have seen https://picovoice.ai/. They have a cool demo video: https://picovoice.ai/products/cheetah/.

But all of their stuff seems to be behind a “contact us” style sales system, so probably $$$.

Rohan

Pocketsphinx works with melodic. What is your issue with it?

Besides, there is also kaldi

1 Like

snips.ai from Paris were definitely the new kid on the blog. Offline and anonymous.


But they got bought up by sonos lately. Their opensource repos are still getting updated:

And here are some docs about how to use it:
https://snips-nlu.readthedocs.io

I met a few guys that were fascinated by it but never had the chance to give it a spin for myself…

Use Rhasspy https://rhasspy.readthedocs.io/en/latest/

{
“text”: “turn on the light”,
“intent”: {
“name”: “LightState”
},
“slots”: {
“state”: “on”
}
}

& same author project - https://voice2json.org/

1 Like

Please, Look https://github.com/MycroftAI/mycroft-core too
Its thing may work offline too

If I remember correctly, Google Voice on Android reverts to local operation when there is no network. When I tested it, it was not as good as the network version, but it was very usable.

1 Like

The only word that works with no network is ‘help.’

Hi, can you use this:

https://rhasspy.readthedocs.io/en/latest/

I’m pretty sure it worked for me–this was on Android 8, IIRC. There might be something like turning voice recognition on in Android that you have to do. I’ll revisit it and see if I can prod my memory.

I just tested my own app with no network–that is, in airplane mode. It did recognize voice dictation correctly. If it would help I will be glad to show you the code. It was interesting that when in airplane the phone would recognize “help” and “OK Google” but it would not recognize “how much is 2 + 2”–it just said that there is no internet connection. However, as I say, my app (Rename Pics by Voice) works.