Training data about genie-server HOT 7 CLOSED

Is you model on Huggingface?

from genie-server.

Hi! Let me answer your questions point by point:

The main training data (semantic parsing aka text to meaning) is synthesized using the genie-toolkit. You can find instructions on how to generate data and train a model in the thingpedia-common-devices repository. I recommend you read our Genie paper first though, as that explains how we synthesize data and build a semantic parser. More recent papers (Schema2QA, SKIM) also expand on the concept for question answering and dialogues.

Snowboy is only used for wake-word detection. We do not use any data or code from Mycroft at the model.

Our model is not on Huggingface at the moment. You can download it from our releases page.

from genie-server.

Thank you for you interesting replies. I'm on Pop!_OS 21.04 and Podman steps on Docker. I was able to build from source, but I backed out using Timeshift because the build is flawed for Debian based distributions. I suppose I can create a Fedora virtual machine and do the install. Maybe I'll do that. I'm glad you aren't using MyCroft. Somewhere I saw Mycroft. I think in one of your code files. It's probably just left over from an old version. I'll read your paper and reexamine your project. It looks interesting. Cheers.

from genie-server.

Yes, we have a different backend for the wake-word that uses mycroft instead of Snowboy. That backend is not currently used because there is no high-quality wake-word model we can use.

from genie-server.

I use vosk. It works very well. Have you tried it? Here's my virtual assistant repo on Github. I haven't worked on it lately because I've been researching tech. I've been thinking about making an intent model, but if I understand what you've done, it seems like a better idea. I've also been looking for a model that translates from English to Sparql. Sparql is powerful, but it's a pain, and a transformer seems perfect for that sort of sequence to sequence task. https://github.com/MikeyBeez/Julie-Julie

from genie-server.

https://www.youtube.com/watch?v=Kky-mdzYLq4&t=489s

from genie-server.

Hope you've had your questions answered, please check out our forum if you'd like to continue the discussion.

from genie-server.