3 ms·

Something I've been watching that might interest you is Mycroft (https://mycroft.ai/ https://mycroft.ai/). It's free software and open hardware version of the e

Something I've been watching that might interest you is Mycroft (https://mycroft.ai/ https://mycroft.ai/). It's free software and open hardware version of the echo, etc.

defo_nonconvex 9y ago
Needs to phone MyCroft homebase for setup (didn't look hard enough to figure a way around that), and uses Google's STT. Privacy...not attained. I shouldn't have to ask Google to open a local file for me.
- khedoros1 9y ago
  > Needs to phone MyCroft homebase for setup (didn't look hard enough to figure a way around that) I think part of their documentation said that you can skip that part, if you want. Ah, found it. Their github Readme has a section titled "Using Mycroft Without Home": https://github.com/MycroftAI/mycroft-core#using-mycroft-without-home https://github.com/MycroftAI/mycroft-core#using-mycroft-with... Apparently, various groups are using some open speech engines (CMUSphinx and Kaldi) to work on internationalization support: https://community.mycroft.ai/t/are-there-any-plans-to-move-from-google-stt-to-something-like-kaldi-or-cmu-sphinx-while-openstt-is-in-development/1494/4 https://community.mycroft.ai/t/are-there-any-plans-to-move-f... > Privacy...not attained. > I shouldn't have to ask Google to open a local file for me. I agree.
khedoros1 9y ago
In a similar vein, I've kept track of Jasper (http://jasperproject.github.io http://jasperproject.github.io) in the past. Conceptually, I like it because it wouldn't require an internet connection (and attendant information leakage). The downsides that I see are that PocketSphinx usually needs some serious tweaking to work well, and Julius certainly needs some pretty heavy configuration to get going. Mycroft seems to be oriented around using their account system by default, and it looks like it's built around Google STT too. I feel like I'd need to do a code audit...or more likely find a way to replace the STT system. Almost 20 years ago, I had a mostly-working speech recognition system on my 400MHZ PC. We ought to have enough functionality in open software now to built a decent assistant without turning to software outside of our control.

Something I've been watching that might interest you is Mycroft (https://mycroft.ai/ https://mycroft.ai/). It's free software and open hardware version of the echo, etc.

defo_nonconvex 9y ago

Needs to phone MyCroft homebase for setup (didn't look hard enough to figure a way around that), and uses Google's STT. Privacy...not attained. I shouldn't have to ask Google to open a local file for me.

khedoros1 9y ago
> Needs to phone MyCroft homebase for setup (didn't look hard enough to figure a way around that) I think part of their documentation said that you can skip that part, if you want. Ah, found it. Their github Readme has a section titled "Using Mycroft Without Home": https://github.com/MycroftAI/mycroft-core#using-mycroft-without-home https://github.com/MycroftAI/mycroft-core#using-mycroft-with... Apparently, various groups are using some open speech engines (CMUSphinx and Kaldi) to work on internationalization support: https://community.mycroft.ai/t/are-there-any-plans-to-move-from-google-stt-to-something-like-kaldi-or-cmu-sphinx-while-openstt-is-in-development/1494/4 https://community.mycroft.ai/t/are-there-any-plans-to-move-f... > Privacy...not attained. > I shouldn't have to ask Google to open a local file for me. I agree.

khedoros1 9y ago

In a similar vein, I've kept track of Jasper (http://jasperproject.github.io http://jasperproject.github.io) in the past. Conceptually, I like it because it wouldn't require an internet connection (and attendant information leakage). The downsides that I see are that PocketSphinx usually needs some serious tweaking to work well, and Julius certainly needs some pretty heavy configuration to get going. Mycroft seems to be oriented around using their account system by default, and it looks like it's built around Google STT too. I feel like I'd need to do a code audit...or more likely find a way to replace the STT system. Almost 20 years ago, I had a mostly-working speech recognition system on my 400MHZ PC. We ought to have enough functionality in open software now to built a decent assistant without turning to software outside of our control.