Mycroft II Arrived - a few small issues!

I’m not sure if these are ‘bugs’ as such but:

  • I live in the UK. Mycroft thinks it’s location is UK - but the time zone is set as GMT-5h. I can’t seem to persuade it that the time zone should be GMT.

  • It will play BBC News, but I only get World Service, not UK News - maybe that’s the same issue?

  • If I ask it to play a song - I only get ones from Russia / Eastern Europe.

  • When I tried to connect to Spotify, it just says there was an error! Likewise for Pandora. I’m guessing this could be a region issue too?

  • I have a slightly Scottish Accent - and am really struggling to get Mycroft to understand me!
    “Hay Mycroft, Install Pandora” - “I don’t know a skill named Taxidermy”
    If you’ve ever watched this sketch, it’s what I’m getting with Mycroft:
    Voice recognition lift (courtesy BBC Scotland) with subtitles.... - YouTube
    It is a serious issue though - that most voice assistants have addressed.

  • Lastly, the “British Male” voice - sounds like an American Male to me. That’s OK - it’s a nice accent, but not what I hoped for. That could be that the voice thinks I’m in the US as well as the clock?

I’ve given up on it for now - and it’s gone for a long sleep in my desk drawer.

I can only give you answers to some of these, but here goes:

  • Re: TZ - Have you set the time zone through the web interface? It won’t auto-detect.
  • Re: BBC - All Mycroft functionality, both default and third-party, lives in a kind of plugin called a Skill. The default news skill only ships with a handful of streams baked in. I haven’t checked, but I think the one labeled “BBC” is the World Service. To get a wider variety of radio stations, one needs a full-featured radio skill.
  • Re: Spotify - Spotify cut off Mycroft’s API access some time ago and they are refusing to reply either to MycroftAI or to any of us, when we harass them on Twitter. The polite customer service person just tells us they’ll pass our comments along. In other words, Spotify is doing this and there’s nothing anyone can do about it until they fix it.
  • How well STT understands you depends on what STT engine you’re using. If you haven’t made any changes, it may well be Google’s speech model failing you (yes, it is; no, it can’t) so my next question becomes, how’s your experience on Android been?
  • Re: “British Male” voice - to clarify, are you sure you were hearing the British Male voice? It sounds very robotic. It’s a whole technological generation behind the American Male. Broadly speaking, if it sounded like a human being, it probably wasn’t the British Male voice.

As for expectations, well, this is the SDK.

Thanks for the quick response.

I set the location during the setup - but didn’t see a time zone, specifically mentioned. Does the Mycroft have it’s own web interface? Most of the mentions relate to using Linux.

Android, on an S10 seems to work pretty well. I’ve just tried the same “Hay Mycroft, Install Pandora” dictating an email - and it got it right. Android does struggle with “Eleven” - consistently translating it to “Unleavened”.

The British Male doesn’t sound like a robot - so I guess it’s not. It’s what I specified, but maybe it was doing me a favour!

I didn’t appreciate this was an SDK. I’m a Hardware Engineer - who just tinkers with software when required. Most of this is way out of my depth. I’m very pleased to have one - don’t get me wrong. It has huge possibilities. It’s just not reached a state of development yet where it’s useful to me. I’ll still have a play with enclosure designs plus the accoustics - and wait for the software to develop.

Yep, I meant setup, poor choice of words. There are, however, some device settings in the web interface where you paired the device. It might be worth a look to see if the settings in there correspond with your expectations from setup.

It’s likely some of these things were bugs that they might not know about! That’s the other purpose of the dev kit period, this is the first big round of public testing (by definition.) Half of what I’ve done with mine so far is just looking for ways to make it crash =P

So it’s definitely worth a look to see if the web interface has the right time zone, voice, etc. If something got messed up, or reset, or failed to sync, that could well be news.

As far as the accent goes, fair enough. STT systems - and Mycroft is compatible with a few, I’m just making assumptions about default configs and which one you’re using - are reliant on a huge corpus of samples. In the not-very-long run, I reckon your best bet will be Mozilla’s STT system, which you’d have to self-host. Mozilla has crowdsourced the acquisition of samples.

Indeed, if you “donate your voice,” you’ll help the world get one step closer to 100% open source STT that works for Scottish accents. It’s not just accents, either. Many of these systems have more trouble with adult women and all children than they have with adult men, simply because there are a whole lot more adult men in most sample sets. It’s a problem, and solving it’s a slow burn. To the extent that the billion-dollar voice assistants have solved it, well… $$ —> voice recordings —> $$ —> LOTS of ML training very quickly —> better accent support than Mozilla’s offerings, or (probably) Google’s public offerings.

Apparently the Mk II is currently configured to override the voice you selected, so that explains the American/British thing. This info culled from here. This is not permanent, but it’s the current situation while some changes are happening to the way configuration is handled by Mycroft-core itself.

So that explains that one.

I have configured a mk2 to use moz tts, and it does work. I haven’t tried any of the mimic1 voices, though.

Good find!
I guess it’s early days in the development. I look forward to seeing how things develop.

Is there a way to make it access my own music collection? At the moment, it’s stored on a Synology NAS. That would be more useful than Spotify or any of the other streaming services.

I’d ask around in chat. I don’t have it mess with media too often. I’d be surprised if no third-party skill exists, though, if nothing else.

Hey if you’re running a Synology NAS with your music then you can also run something like Emby on the same NAS then use the Emby Skill.

We are working on providing some default music Skills - as Chance mentioned this has been hampered by Spotify being less than helpful, but also many services specifically say you can’t use their service with a voice assistant all whilst letting Google and Amazon do exactly that. Anti-competitive much? One way or another we will find a solution.

Pandora is US only as far as I’m aware. I’m in Australia and they shutdown here years ago.

I’m curious about the TZ issue - can you DM me the location you set for the device? It might be that our database has associated it with some other city of the same name or something else entirely.

As mentioned, the voice config is currently limited to the US male - have a fix in the works, just working through some finer details on implications of the change.

For the speech recognition we’ll need to look at alternative options and see what works best for different accents. You can configure this manually through the command line mycroft configuration manager, but this is something that will also be provided as a selectable option in It may also be worth us suggesting a specific configuration based on the country your device is registered in as there are models specifically designed for certain accents. Currently we use the same API / model for all users.

Things are progressing quickly here though, so keep an eye on the Mark II updates as we tick these things off :slight_smile:

Thanks Gez, and Chance.
I appreciate your taking the time.


1 Like