Hacker News new | ask | show | jobs
by juriansluiman 2276 days ago
Mycroft as software is used by a small group of users and seems pretty stable. More features are continuously added and the design principles look promising (open source, as private as possible).

The biggest problem is their hardware: they have a Mycroft v1, (to me personally) a prototype alike piece of hardware. There have been successful campaigns for a v2 release, with new hardware and an improved design.

However, they fail to work with reliable partners and there's still no working device which resembles the final production level. I have been a backer of the indigogo campaign but it's frustrating they postpone their Mycroft v2 every time again. I really hope the can deliver the device at some point, but they keep rewriting software and if they ship, the hardware is pretty outdated probably.
2 comments
tekchip 2276 days ago
It's true v1 was more of a proof of concept device. The core hardware target has always been the raspberry pi family of devices since they are pretty well ubiquitous. 2-3 is the current set. Base software is linux so if your audio devices, mic and speaker, work with Linux then you're pretty well good to go. Most microphone hats for Pi are supported. I use the Google Voicekit AIY v1 with raspberry pi 3 b+. Works a treat.

As with most open source efforts(especially early on) there's a lot of tinkering and DIY at the get go and they've designed their product to be supportive of this. Their "retail" devices are, much like Googles intent with Nexus, intended to be a best possible target for other vendors to target including the DIY crowd. Whether that's the right way to come at it is open to debate but the premise that they don't have a set hardware target is at best misleading.

Edit: Link to hardware spec https://mycroft-ai.gitbook.io/docs/using-mycroft-ai/get-mycr...

link
bronco21016 2276 days ago
I think perhaps it’s hard to hit a hardware target when the software is still in a pretty big state of change. It’s hard to say what hardware will be needed and what is the best compromise between hardware performance and the DeepSpeech NN for STT. DeepSpeech is still in development as well.

I think the priority needs to be getting the STT to a good, neural net backed, open source engine. Once the software is stabilized I think there’s room for a whole ecosystem of hardware interfaces.

link