Last little teaser for now. Really need to get back to work. 
The team is working hard on the new audio system, that in the end will look like this;
2 Likes
Wow, very cool! What is the skill that plays the music? Thanks.
-Mike MOne of the cool things about UI handlers in Mycroft is that any media skill can bring up those controls.
@mike99mac That is the simple-youtube skill from @JarbasAl We decided to include it by default within the image (till we got told not to 
)
But yeah, the stream comes from youtube, but if I install deezer/spotify/bandcamp all together, it just plays the best match and you don’t know where it comes from until you swipe the screen to select something different.
Below another video of the first run wizard, to configure your OpenVoiceOS device the Mycroft way.
3 Likes
the common play UI is now all plugged in and going strong, repeat/shuffle buttons are not yet integrated however
I’m sure there will be a proper demo video shortly
2 Likes
The OpenVoiceOS team is pleased to announce Precise Lite prealpha, and a companion plugin! This is Precise on TFLite, presenting a much smaller footprint. It’s suitable for featherweight devices, and we also plan to ship it with OVOS itself.
A modest amount of setup is required; in addition to installing the wake word engine or plugin, you’ll need to configure Mycroft to use it, and point it at a model. See below for an example config snippet.
Like most Mycroft plugins, the Precise Lite plugin is believed to be compatible with any up-to-date Mycroft device. Mark II, Mycroft-git, Chatterbox, Plasma Bigscreen, you should all be able to use the plugin. I know there are a lot of us running 2GB Pi-based devices. It’s a real boon for us!
The primary author on this one was @jarbas. Special thanks to: Bartmoss St. Clair and eltocino.
To test, in your Mycroft environment,
mycroft-pip install git+https://github.com/OpenVoiceOS/ovos-ww-plugin-precise-lite
(Or if you run it systemwide / on a release of OpenVoiceOS)
pip install git+https://github.com/OpenVoiceOS/ovos-ww-plugin-precise-lite
and, assuming you’ve cloned the models repo into ~/precise-lite-models, see the following:
# mycroft.conf
"hotwords": {
"hey mycroft": {
"module": "ovos-precise-lite",
"model": "~/precise-lite-models/wakewords/en/hey_mycroft.tflite",
"sensitivity": 0.5,
"trigger_level": 3
...
Like mentioned above, the next pre-alpha image of OpenVoiceOS will ship/default configured with this precise-lite version of the wakeword listener. We are also currently looking into adding the “wake up” wake-word as secondairy precise-lite wake word to replace the pocketsphinx one, however we might reach out to you all to help out in gathering the required audio data to train an initial first version of it.
Other related news; The next pre-release will have a major bump of all required packages;
- Based on buildroot 2021.02.05 LTS
- Latest linux kernel 5.10.60-v8 LTS
- Latest XMOS VocalFusion kernel driver for linux 5.10
- KDE Framework (KF5) bumped to version 5.82 (kf5-kio / kf5-kirigami2)
- About any other package bumped and or latest and greatest.
It will also have some not yet finished goodies such as a local configuration webbased dashboard;
The video GUI player has been worked on, to follow the same appearance as the above showed music GUI screens
and to top it up, the first run wizard - local selection procedure is now finished / implemented, providing a very easy and simple way to run Mycroft fully local and/ or without internet.
3 Likes
The included weather skill with animated background based on the weather report and time of day (dark/light mode).
And yes, typical Dutch weather…
2 Likes
Instant listen feature that @JarbasAl worked on and is default within OpenVoiceOS.
In short it means it basically immediately start listening for the intent when it hears the wake word which includes the sound that is actually being recorded as the wake word.
And as the SJ201 hat of the Mark2 is such a great little device, even when we do not disable the listening sound (although we made it a bit less intrusive and shorter in this video) it perfectly understands what you saying as all playing sound is removed by the DSP.
(The above weather skill demo already kind of gave it away, although with the “longer” listening sound)
3 Likes
Feature or Bug? Should he say second timer when mycroft opened the third? 
And to your Video: Awesome! I love to see your progress! Nice work.
Report that to the Mycroft A.I. chaps 
It is a bit confusing but it is indeed the second timer for 5 minutes
actually a crude feature. the id is sorted by timer length (if unnamed).
Nice one - weather is looking slick and the instant listen is a great improvement.
In terms of the Timer behaviour - definitely interested in feedback. We’ve gone with more verbose responses to start with because it can then get confusing if you say something like “cancel the 5 minute timer”, or when the first 5 minute timer expires and it says “your 5 minute timer is up”. So the response on the “rice” here, is to help remind the user that there are actually two 5 minute timers.
Sorting by timer length is a newer thing too. Again keen on feedback of whether it’s helpful, confusing, or you don’t care as long as it’s consistent?
Installing ``ovos-precise-lite on a picroft went pretty smoothly and it was notably more responsive to the wakeword although I still had to wait until the beep to be able to issue commands.
The problem I now have is that mycroft.client.speech is now taking around 175% cpu compared to 40% for precise-engine and 100% for pocketsphinx, I did a full reboot to be sure and got the same cpu usage.
I did get t```he error:
ImportError: liblapack.so.3: cannot open shared object file: No such file or directory
on my picrofy install meant I needed to install libatlas3-base and then reinstall numpy:
sudo apt install libatlas3-base
~/mycroft-core/bin/mycroft-pip install --upgrade numpy
I also needed to add
"phonemes": "HH EY . M AY K R AO F T",
to the mycroft.conf (as per the precise and pocketsphinx sections).
After this and a $HOME/mycroft-core/start-mycroft.sh voice restart Mycroft worked. I think that it’s worth mentioing the package dependency in the install instructions.
1 Like
Thanks for that and sound very plausible as I seen other similar reports about the need of (re)installing numpy AFTER lapack.
@AIIX I think you are very interested in this as you had the same high CPU issue.
1 Like
I am really interested in both the seamless “Hey Mycroft do something” solution and probably the dsp processing you mentioned too (I use snapcast to play audio on the mycroft device and may need it to process that sound source if it’s not happening automatically).
Also would it be best to discuss the CPU load here or as an issue on github?
I notice other things like failing to connect to PA and comments about Threshold missing (cc @AIIX)
The issue I am see with the CPU usage may not be related as I believe the picroft image is Arm64, I am not quite sure yet but you can run a “uname -a” to check, The platform I am trying to get the precise lite runner working on is the armhf architecture running python 3.8 on ubuntu focal, I have already tried installing libatlas3-base and building both numpy and scipy from scratch but it didn’t seem to help, whereas I have no CPU usage issues on Arm64 following the above guide using the same python and distro combination
I may have used “picroft” misleadingly. I’m running vanilla rasperry pi OS 10 (lite) with mycroft installed on top.
It has the 32bit armv7l kernel
RpiOS is arm-linux-gnueabihf
python is 3.7.3
The seamless solution you mention is called; “Instant listening”, which is not (yet) part of mycroft-core but the code changes are here;
1 Like
ah right - so it “just” starts listening at once - I thought that it may be integrated into precise so that precise kinda stopped overwriting its ringbuffer when it was activated and streamed the post-wakeword sound to be STT’ed
I guess there is no need for that if the wake recognition → start recording latency is low enough
(and I also suspect that the dsp remove-any-playing-sound feature would help improve STT if it was active?)
precise kinda stopped overwriting its ringbuffer when it was activated and streamed…
The day after it came up, @gez-mycroft and I almost simultaneously arrived at (basically) the same solution in different channels on MycroftAI’s chat =P
That’s a core change, though, as it would require 1) switching to a single, persistent input stream, 2) adding a secondary buffer, so the WW can put a few frames back before the handoff, and 3) rewriting the Mycroft-side routine that handles the listen/record/stop cycle.
A good proposal, and we should all run with it, but it’s more than can be done in a plugin.
3 Likes