Build an open future with us.

Invest in Mycroft and become a community partner.

The Machine Learning Team


#1

Hi all,

The OpenSTT project and Mycroft’s XPrize engagement have just been announced and a lot of amazing work has to be done. Therefore, I’d love to start knowing you better.

Please comment on this topic and share more information about your experience and interest in AI and ML field. Developers are always welcome. There are many frameworks in different languages we have to evaluate. Whether you are an expert or not, you can also help in other ways such as translating texts, manipulating audio and images, talking with AI and ML communities and so on.

Let’s build the Mycroft Machine Learning Team and work on STT and other awesome AI challenges!

I’m looking forward to know more about you.


Introducing the OpenSTT Project
#3

Awesome! What kind of skills are you looking for in developers? I’ve always been very interested in the AI field, but I have no real experience there. I’m an experiences developer (been around for 10 years now), I’ve always worked on the web platform (and IoT hobby projects), mostly with Java but also with Python, Javascript and C# and some PHP.

I’m eager to help though, that should count for something :wink:


#4

Thank you for asking that @mcampo! I’ve edit the post a bit to answer that. Basically, skills in all programming languages are welcome. The number of released ML frameworks has increased a lot in the past months and the list is simply huge. We’ll have to evaluate as many as we can if we want to create a high efficient STT engine.

For instance, TensorFlow from Google requires mainly Python development. Warp-CTC from Baidu was developed in C, there are a bunch of frameworks in Java like H2O and DeepLearning4J, and many others in different languages. So, you are more than welcome!

Very soon, I’ll provide more information on how we’ll work on those. Right now, I need to know as much as possible from the community. Invite your friends and make this post public sharing it. Thx!


#5

Thank you for this topic Mycroft is amazing project our team working for Arabic support in mycroft , I want if we can app in C# for mycroft thank you


#6

Dear all,

thanks for the invitation to join the ML team!

My name is Wolf, and I am currently pursuing a Robotics Master at ETH in Zurich.

A project that I am planning since a while is the OpenBrainInitiative. The goal is to collaborately create a database that basically works like OpenStreetMaps, but for arbitrary data geared towards machine learning. The database would be essentially an API that communicates via changesets, so that for example a better voxforge could be created on top of the OBI (does that make sense?).

I believe you guys are thinking in similar directions so I’d be happy to share my insights to combine knowledge :slightly_smiling:

Cheers,

Wolf


#7

Dear all,

Awesome project!
Got myself a MSc in AI and Robotics back in 2002 and got years of development experience in Java and C and now getting into Python.
Would love to be able to contribute!

Antoine


#8

Hi all,

I’m highly interested in helping out. I have a master’s in bioengineering and have a few courses in machine learning (neural networks, swarm intelligence, other computational intelligence) that I would like to put to use. I’m best at C and C++, but I’ve worked with Java, Python, and some others (dot net mainly) to some extent. Looking forward to seeing where this project goes.

Mike


#9

That’s a very interesting idea. Have you ever heard of OpenML? This project seems very similar to what you thought. Check it out. I intend to try their platform very soon.


#10

@jdorleans we should see if that project wants to partner with us on OpenSTT.


#11

I’m following this project because I want to use it for my Star Wars BB-8 Build: http://sacramentobb8builders.club/

Currently running Python/OpenCV atm.

I also am interested in some fringe elements of AI/ML that I want to incorporate into my side project: www.beaconsafety.co - working on getting some more advanced stuff for the platform.


#12

Hi guys
I know yours asking for developers and programmers with unfortunately I know skills in either but keen to help in other ways ! Don’t have much working knowledge of AI willing to learn all I can and hopefully pick up some programming skills along the way.
This is an amazing opertunity show case the power of the open source community, what are the likely implications for the community as a hole when/if mycroft wins :wink:
Found a use for me! i gave my voice to the mimic project :slightly_smiling:


#13

Hello all, my name is Alexander

I am very interested in helping out. I have about 2 years of experience in various machine learning applications, mostly centered around natural language processing and vision software. I have spent the last year developing vision object detection and localization neural nets for a robotics team. I know TensorFlow very well and I know many methods of training nerual nets for various purposes including NLP, vision and prediction / planning. I know all major programming languages including Java, C++, Python, C, .NET and more.

Aside from machine learning, I am proficient in gentoo, debian and red hat linux. I have done system administration jobs for the past 3 or so years managing small server systems such as websites and forums.

I would love to help in any way I can!


#14

That’s amazing to hear from you guys @Craig_Merry, @Ubuntu_padawan and @Robo. We are on researching phase for our OpenSTT project and you are more than welcome to participate. Until now, we’ve been experimenting with Kaldi and its NN models TED and LibriSpeech. We are also starting to work on other ML projects very soon and you might be interested in joining them as well.

I’ll send an invitation to you all to join our chat platform so we can keep a closer contact.


#15

Hi jdorleans thabks for the invite will have look and see if there is any i could help with


#16

Hi i am very interested in this project, and specifically that it will integrate into Ubuntu phone and desktop. I have a background in making an AIML chatbot in python in ubuntu call butlerbot. the code is here… https://github.com/betzelelgalut i was able to get it to open programs from teh intereactive chat shell in the bash terminal. however, it moved on from that project in light of UIMA, and other projects which caught my interest. i have had some experience with python, and java. mostly i prefer python as it runs leaner than java does, and a bit easier to code for.


#17

Hey, I’m also up for some coding! I’m a C and C++ developer with 4 years experience in Autonomos Cars. Nothing to do with text, but I’m a software architect and I know a thing or two about good code structure.
Looking forward to helping out.


#18

Hi, I have a PhD in Computational Science and Informatics (Computational Statistics), and have developed and extended several machine learning methods through my research. I don’t have a huge amount of free time, but am willing to contribute and evaluate methodology as well as do some coding.

One issue that might be out-of-scope is the intent parser. Right now, it’s just a multinomial likelihood ratio, with some comments about improving it some day. I don’t think this method will get you beyond the “command line / text adventure” approach to voice queries. I have no issue with some sort of likelihood ratio approach, but the keyword parameterization is severely limited beyond just creating a proof-of-concept.

The next time I have some free time I’ll do a bit of a lit review and try to put together an alternative intent parser, probably written in R or C. Is there any documentation on interfaces between the STT -> Intent Parser -> TTS?

Thanks,


#19

Hi Jonathan,
Linagora is an open-source company located in France. We are developing hubl.in (https://hubl.in/) the open-source video conference service. We are working on STT to add new features such as real-time recommendation, automatic summary or subtitles. We are also working on Kaldi and developping models for English, French and Arabic languages. Perhaps we may collaborate ?
By the way I will be in ICML this month in NYC, will you be also ? perhaps we can talk ?
Jean-Pierre
Innovation director at Linagora
@jplorre


#20

@jplorre yes! Let’s see if we can collaborate. I’ll PM you along with @jdorleans.


#21

I’d be interested in helping where I can.

I work as a lead visual designer at Canonical on Ubuntu Touch/Desktop, and in my spare time am playing with machine learning. I’ve got roughly a year of experience with Caffe, currently learning Tensorflow.

I have an Amazon Echo Dot- I think the hardware is great and my kids love it (mostly for jokes and Spotify), but I find the software integrations very limiting. I am also hoping we can come up with accurate speech recognition that runs completely on the device instead of the cloud, on current (or near future) mobile GPUs. Shrinking model sizes with pruning and quantizing weights to 8 bit looks promising for mobile use.We also need better CUDA alternatives for mobile GPUs. Has anyone looked at Vulkan/SPIR-V for deep learning? This implements some basic deep learning on Metal (and their paper has pointers for doing the same with SPIR-V) - http://deeplearningkit.org/

On the hardware side I think we also need readily available array microphones (and OSS drivers that do beamforming) for far-field audio. Has anyone found a source for affordable MEMS array mics?

You might be interested in this - It’s based on Kaldi but has its own CTC implementation