Mycroft voice assistant

[nas]

Anyone tried installing mycroft?

https://mycroft-ai.gitbook.io/docs/using...roft/linux

Tried it on pureos (without mimic tts engine) but mic not being picked up. Will have a look at the troubleshooting below but someone might want to try on mobian in the meanwhile.

https://mycroft-ai.gitbook.io/docs/using...leshooting

[pjsf]

Anyone tried installing mycroft?

https://mycroft-ai.gitbook.io/docs/using...roft/linux

Tried it on pureos (without mimic tts engine) but mic not being picked up. Will have a look at the troubleshooting below but someone might want to try on mobian in the meanwhile.

https://mycroft-ai.gitbook.io/docs/using...leshooting

Had a quick look, but on mobian it wants to remove libjack-jackd2-0 due to conflicts (looks like it'll be replaced by libjack0), given how crucial the sound system is to a phone, I decided against going further.

[nas]

Had a quick look, but on mobian it wants to remove libjack-jackd2-0 due to conflicts (looks like it'll be replaced by libjack0), given how crucial the sound system is to a phone, I decided against going further.

I didn't notice any conflicts on pureos but it may be due to it running debian buster rather than bullseye on mobian. From the troubleshooting guide, it looks like my problem is an audio routing issue on pureos.

Edit: Voice is detected on a mobian install. Created a mycroft account and paired the device. I guess need the text-to-speech engine (mimic) too but the build is failing at the moment.

Extra: Google is the default speech-to-text service but it can be changed. A less worse option may be IBM Watson.


  mycroft-pinephone.jpg (Size: 92.29 KB / Downloads: 542)

[nas]

The mimic1 build was failing due to memory running out as explained here. It would only complete  using one thread/core (make -j1). However, running the mimic command produces no sound (except a low whine) and no error either.

I'm not sure if a local tts engine is necessary though. The docs say it should still work with the remote mimic2 at the server end. More reading required.

Edit: No luck in getting mycroft to speak. It detects "hey mycroft" and even confirms a request to play the news but no sound. I suspect something went wrong in compiling the voices since it completed within about 30mins despite dev posts saying it would take hours. The "make check" passed all tests for mimic though. Running out of ideas so maybe someone else can give it a shot and see if they can get a bit further.

Extra: A potential way forward may be to install a different tts engine as described here.

[vanja]

I'm using mycroft on RPI (picroft) for controling openHAB platform (smart home). What I would like to see is mycroft linux client that just connects to "main" mycroft in the house and forwards him commands.
I guess no need for "full" mycroft installation in that case...but that's up to mycroft developers to develop...

[nas]

Basic local voice recognition with some remote services would be nice to get working on the phone. Pair that with a virtual desktop assistant then geekish dreams could come true.

No luck finding such a project on linux but I'd guess the gizmo in the below link isn't running on high end hardware so should be feasible.

https://www.digitaltrends.com/mobile/gat...a-ai-news/

Edit: By the way, espeak can be installed with apt and its text-to-speech engine works fine. Unfortunately, not able to test mycroft with it since I'm not on mobian at the moment.

[nas]

I tried it again recently and it works - kinda.

The wakeword recognition is instant but then there's a few seconds delay until the confirmation beep. Unfortunately, voice recognition rarely works and usually crashes the microphone input - poss software conflict as mentioned above. Although typed commands work fine and produce spoken replies about the weather or play news audio streams.

Has potential but needs a bit of work from the mycroft developers really.

[ChanceNCounter]

Necro, but I search periodically to tell people not to beat their heads against the wall while community devs work on the "mobile Mycroft" problem, and stumbled on this thread. I think I should add two things for future browsers:


First, Mycroft will never run 100% locally on the PinePhone and don't even try =P The name brand assistants are in the cloud for a reason. The only successful mobile Mycroft "builds" have been thin clients, connected to an instance of Mycroft running elsewhere on your network.



And that's okay. Your memory and battery life will thank you. The important thing is that *you* control that instance of Mycroft, where it's hosted, and what it does.



Second, @vanja what you asked about 1.5 years ago XD is the thin client thing, and something community members are working on, if you haven't discovered it already. Head back to Mycroft places and ask Jarbas about HiveMind. That's what I'm using on my PinePhone (very early days) and it's intended to support a whole-home or whole-PAN mesh. One AI per person, connected to many devices, is the idea. Siri, eat your heart out.

[Lousy Fisherman]

I have been building Kalliope which uses a lot of Mycroft code. Thank you.  It all works, wakeword, voice recognition and STT. There are issues with performance, for example all TTS phrases need to be pre-built and ALL python code needs to be pre-loaded

I expect to have an alpha package build of Kalliope, both bare metal and docker images, integrated with Maps, contacts and phone control by the end of February, 2021. At this time I have a fully working POC on the  Arch and Mobian Pinephone distros.

The main difference between Kalliope and Mycroft is philisophical, there is no central server in the Kalliope world. However, I see no reason why the Mycroft linux client cannot be ported to aarch64 successfully.

YMMV
LF