fossasia/susi_hardware

https://github.com/fossasia/susi_linux

Michael Christen

@Orbiter

@norbusan one of the sensors I have has a wake-up mode: to save power, it only switches on wifi if measure values have changed. Then it contacts a server endpoint and transmits values. We could add to our flas server such an endpoint for IoT devices with such a wake-up mode.

Norbert Preining

@norbusan

@Orbiter Sure enough. Can you flesh out the API requirements and what should happen, and make an issue to susi_installer repo?

Mario Behling

@mariobehling

@/all Dear friends, despite challenges we are happy to work with the community to share live streams from the FOSSASIA Summit "social distancing event" and include remote speakers and participation. An overview of streams is here: https://www.youtube.com/fossasiaorg
The complete schedule of the event is here: https://summit.fossasia.org/event/schedule.html

Michael Christen

@Orbiter

@norbusan sending a default skill to the etherpad should be done by the server itself because that works then also in a minimal desktop setup.

I will work on that.

I remember that there is already a check to identify a non-customized etherpad

So I can easily build on that.

Norbert Preining

@norbusan

@Orbiter fine with me, can you let me know how long this will take? I will create a new image for testing when this is done.

Michael Christen

@Orbiter

the next days

Norbert Preining

@norbusan

Thanks, please send me an email when done!

Nguyễn Hồng Quân

@hongquan

Sorry for disappear for so long.
I've just heard about this: https://github.com/Picovoice/Porcupine

Wake word alternative to Snowboy.

This page https://rhasspy.readthedocs.io/en/latest/wake-word/ rates Porcupine highly.

Norbert Preining

@norbusan

It is even worse license-wise then snowboy:

This repository is licensed under Apache 2.0 which allows running the engine on all 
supported platforms (except microcontrollers) using a set of freely-available models. 
You may create custom wake-word models using Picovoice Console for non-commercial
and personal use free of charge. The free-tier only allows model training for x86_64 
(Linux, Mac, and Windows).

So the free version cannot be used on Raspi, etc etc etc ...

Nguyễn Hồng Quân

@hongquan

Thanks for pointing out.

Norbert Preining

@norbusan

:-)

No prob. Unfortunately there are lots of hotword systems, but none really is open source.

There is one ... I forgot the name. .. Julius or so, that is fully open.

Nguyễn Hồng Quân

@hongquan

Nice

Norbert Preining

@norbusan

I am just trying out the susi build based on the new raspian os released a few days ago. Seems to be rather similar. Will see.

Nguyễn Hồng Quân

@hongquan

I also have just order one RPi 4. Will help test.

Norbert Preining

@norbusan

Great. Thanks. The current builds run completely offline, using deepspeech for STT and flite for TTS. DeepSpeech isn't really great, though.

There are a few things that need to be done, most importantly porting my extension of SpeechRecognition library with DeepSpeech support from the normal API to the deepspeech streaming api to get better speeds. Need to think about how to do this. If you have interest in looking into this, let me know, I can show you where to start.

Nguyễn Hồng Quân

@hongquan

Oh, it changes a lot since my last time involvement

Norbert Preining

@norbusan

What do you mean with "it"?

Well yes, the state machine has been rewritten because it was not reentry save, with multiple thread there were problems. Furthermore, the call stack got deeper and deeper over time, long running susi would hang at some point. And it is a bit more simple and easy to understand now ;-)

Other than this not sooo much has changed, besides the devops part on how to make images, etc etc. And that the installer is now working also on the desktop.

Nguyễn Hồng Quân

@hongquan

"it", I mean the stack of "susi linux". Last time, DeepSpeech was not included.

Norbert Preining

@norbusan

Ah yeah, that was simply an addition to SpeechRecognition module I implemented, to allow for STT on device.

Mario Behling

@mariobehling

@norbusan how about the Mycroft hotword component? You mentioned in the past this could be adapted after a few changes maybe.

Norbert Preining

@norbusan

@mariobehling the problem is including it into SpeechRecognition library, which does all the heavy lifting for us till now.

Mario Behling

@mariobehling

SUSI.AI weekly meeting taking place now: https://meet.jit.si/SUSI.AI

Mario Behling

@mariobehling

@hongquan Here is a link to the Julius. What do you think? https://github.com/julius-speech/julius

Mario Behling

@mariobehling

@norbusan @Orbiter In the meeting you mentioned we would not necessarily need the setup of the device anymore to run it locally. It is correct that it can run without the Internet, but I think we would still want to setup Wifi and provide options to change the STT/TTS settings. So, would be good to keep the setup.

Norbert Preining

@norbusan

@mariobehling I have updated the susi installer so that:

susi linux (the voice component) is started automatically immediately, since we are now defaulting to all local (flite/deepspeech)
the setup page is still available, of course, for those wanting to setup wifi and other sst/tts

@Orbiter the lastest changes also put etherpad into susi_server/data/etherpad-lite
You used the hard-coded 1.8.0 version, while the code I use in the susi installer grabs the latest release tarball (currently 1.8.4) and uses it.

Michael Christen

@Orbiter

great! Yes I recognized that your version visually differs from the one I was using.

Having the ability to use voice before setup is also a great proof that we can operate completely offline!

Have you seen the chat protocol in the etherpad chat? (works only on the susi page, did I mention this?)

Norbert Preining

@norbusan

Still building images (first try failed).

What do you mean with "works only on the susi page"?

Michael Christen

@Orbiter

The SUSI pad

Michael Christen

@Orbiter

etherpad

Nguyễn Hồng Quân

@hongquan

Recently I tried GStreamer to build a personal tool, and found that, it defines a structure to let processe simultaneously multimedia data very well: https://gstreamer.freedesktop.org/documentation/application-development/introduction/basics.html?gi-language=c#bins-and-pipelines

I think we can apply it to solve the problem of concurrency (listen to hotword when playing music) .

My use case in the personal tool is that, I get the video stream from webcam, split it to two lines, one is to display in my app window and one to scan for the QR code in the video image.

Norbert Preining

@norbusan

@hongquan that problem is already solved ;-) We can already play music and listen to hotwords etc.

Nguyễn Hồng Quân

@hongquan

Nice!

Mario Behling

@mariobehling

FOSSASIA OpnTec meeting for hardware projects, PSLab, Open Spectrometer etc. is starting in a moment https://meet.jit.si/fossasia-hw

Mario Behling

@mariobehling

Please join the SUSI.AI desktop installation workshop. It starts in a few minutes here https://meet.jit.si/SUSI.AI
This session covers step-by-step guide how to set up and configure Susi on Linux

Mario Behling

@mariobehling

FOSSASIA Open Hardware weekly meeting starting now https://meet.jit.si/fossasia-hw

Michael Christen

@Orbiter

A new susisbian image is available at http://download.susi.ai/susibian/ susibian-release-20200712.0.img.xz

Mario Behling

@mariobehling

Dear @/all, we are running the next version of our SUSI.AI installation workshop. This workshop is a run through to test our installation process for SUSI.AI on smart speakers. Compared to the last time we have improved a number of steps. The goal of this workshop is to make installations of SUSI.AI on smart speakers easier and enable more contributors to run workshops independently in future. Please join us at 8PM SGT, 5.30PM IST 2PM CET here https://meet.jit.si/SUSI.AI
(in a few minutes)

Mario Behling

@mariobehling

@/all We are running a PSLab installation workshop with Daniel Maslowski. We are starting in a few minutes. In this workshop you will learn how to install the PSLab Desktop and Python app, connect the device and test its functionality.
Please join the workshop here: https://meet.jit.si/fossasia-hw

More info about PSLab at https://pslab.io
PSLab Desktop https://github.com/fossasia/pslab-desktop
PSLab Python https://github.com/fossasia/pslab-python

Mario Behling

@mariobehling

Dear @/all, the next SUSI.AI workshop is a run through to test our installation process for on the desktop. The main goal of this workshop is to make installations of SUSI.AI on the desktop easier.
Please join us here https://meet.jit.si/SUSI.AI on
Tuesday, August 4, 2020
15:00 Berlin | 18:30 India | 21:00 Singapore | Tokyo 22:00

Where communities thrive

People

Repo info

Activity