The Alysha Linguistic AI Project

Here just to express my happiness seeing you back here. And to ask you about your health.

For the project, I find it interesting and something of great help. AI is always fascinating subject. And i am hooked as I read this -

Knows 50 to 100 words. Uses short, two- or three-word sentences and personal pronouns ("I falldown!" "Me go school?").

Very interested in this. I've been trying to get good speech in and out of Linux for quite some time and it's not easy. Please post updates 🙂

Posted by: @supercharlie

Very interested in this. I've been trying to get good speech in and out of Linux for quite some time and it's not easy. Please post updates 🙂

Hi Charlie I'll be doing an update on this in a little bit. The speech recognition I'm using is CMU pocketsphinx it's not the best and there are far better ones but I'm using this one for specific purpose in my project.

The speech recognition and text to speech the Epseak are used for general input output of the words. Technically they are not actually part of the linguistic AI system.  This is just the method for getting the words in and out. I could have used just keyboard input output but since I'll be working on this for several years I thought it would be nice to have the speech recognition and have the robot actually talk.

I'm told there are far better speech recognition engines including one called Vosk that is independent from the the web, and several others that require clouds like Google or so forth.  I prefer to keep things simple for this linguistic AI project.  So this is why i went with pocketsphinx.

@robo-pi

You are speaking my language.

I've had an off and on project to animate a furby for about three years.. I came to the PocketSphinx solution as well. I tried vosk about a month ago and it was super easy to install but the accuracy was about half of PocketSphinx and didn't have an easy way to do keywords that I saw.

The text to speech solution I came up with was Flite.. festival lite I believe. The RMS voice is about as good as it gets local. Full festival works but it's difficult to change the voices.

And I completely understand that those are merely the in and out to the processing and that is where the magic is 🙂

Looking forward to the updates 🙂

Onward and upward!!!

Posted by: @supercharlie

Looking forward to the updates

I have some play lists on my YouTube channel you might want to look at.

https://www.youtube.com/channel/UC0Kfb4KDLvKzg_CQe8IYooQ/playlists

@robo-pi

I had no idea  espeak was so flexible. I really like that you can change it on the fly by manipulating the config file.

I think Flite sounds just a smiiidge better and since I'm already embedded with it I'll keep it but eventually when it gets down to manipulating the voice by emotions or whatever it's looks like espeak is the only game.

Nice video intro too 🙂

Posted by: @supercharlie

I had no idea  espeak was so flexible. I really like that you can change it on the fly by manipulating the config file.

I'm really glad I did the research on espeak as it did turn out to be a very nice program for my purpose with a Raspberry Pi 4.   But as far as the linguistic AI program is concerned the speech synthesizer is not really a part of the linguistic AI program directly.   It simply takes in speech and turns it into words that have no definitions so it's not part of the linguistic AI program.   It's just for input/output of undefined words and their phonemes.

@robo-pi

Does espeak support CMS Arctic voices? I really like the RMS one I think they're also called Clunits.. or just mbrolla?

@robo-pi

Nice project!  I used a DIY Google assistant / voice kit so that I could move a robot manipulator verbally.  I am going to look at PocketSphinx, sounds like a great tool!  I have never been a fan of using Google's cloud api, it was just an easy way past that requirement.

As a bit of an aside, I also love your project name.  I have had a 2 year project going on called "Semantic I/O" for me, the name was inspired by Berners-Lee's semantic web.  My hope there is to build an ontology of human-computer interactions as well as computer-computer interactions so that the communication protocol can actually act as the primary abstraction layer for IoT devices.  More here if you are curious: https://semanticio.org/  

Your document leads me to believe your goal is for the robot to be understand / respond to speech and "learn" (new words? Comprehension leading to action of some sort?) similar to how a child does?

Some ideas in this area I hope may be helpful, research on childhood development (I have personally been looking at this one: Early Category and Concept Development: Making Sense of the Blooming, Buzzing Confusion, David H. Rakison)

Also less actionable, but fun to read in this area is Jeff Hawkins "On Intelligence"

Posted by: @supercharlie

Does espeak support CMS Arctic voices? I really like the RMS one I think they're also called Clunits.. or just mbrolla?

I am currently using the mbrolla voices with espeak.  However it's important to keep in mind that this speech recognition engine, "pocketsphinx" and the text-to-speech synthesizer "espeak" are really separate from the linguistic AI part of the programming.   Those utilities can be changed at anytime.  They have nothing to do with defining the words in this linguistic AI project.

It is important to understand this distinction because if you get lost going down the speech synthesizer and speech recognition parts you're not doing the linguistic AI.

Posted by: @briang

Nice project!  I used a DIY Google assistant / voice kit so that I could move a robot manipulator verbally.  I am going to look at PocketSphinx, sounds like a great tool!  I have never been a fan of using Google's cloud api, it was just an easy way past that requirement.

Yes there are a lot of applications that use speech recognition in various ways what you were speaking of above using speech recognition in a sense of an expert system where you decide what the words mean and you program your robot to react to them in this way.

When doing this many times the robot doesn't even need to know the meaning of the words since you are defining the actions the robot will take when it hears these words.

Posted by: @briang

Your document leads me to believe your goal is for the robot to be understand / respond to speech and "learn" (new words? Comprehension leading to action of some sort?) similar to how a child does?

Yes, absolutely, that is the heart of this linguistic AI project.

It is based in part on Marvin Minsky's "society of mind" where there are agencies that these dictionaries will become, and it is also based upon human child development so that the robot learns in a very similar way that a human child learns.   That is the concept of the Linguistic AI project.

Posted by: @briang

I have had a 2 year project going on called "Semantic I/O" for me

I've moved away from using the word "semantic" to using the word "linguistic".  The reason I've done this is because commercial AI companies are using the term "semantic AI" for different purposes entirely.  They are using deep neural networks to make connections between various meanings of related words usually for the purpose of helping advertisers find consumers for their products.  That's certainly a valid form of artificial intelligence it's just not the same as the Linguistic AI I am attempting to create here.

@robo-pi

Just looking for tools for my tool bag 🙂

Your project is awesome and I realize the TTS and STT are just the input and output devices. It's what's in between them that matter 😁

Posted by: @supercharlie

Your project is awesome and I realize the TTS and STT are just the input and output devices. It's what's in between them that matter

Absolutely and there's a lot you can do with these tools.  I'm using Microsoft dictation for my posts and my word documents.  And this is also a speech recognition engine that returns my speech into the words you see on the screen right now.   It does a very nice job it has a very large vocabulary, and is very impressive indeed.    But of course it wouldn't be good for my linguistic AI project as I'm more interested in defining individual words one on one for that project.

Just posting a quick PDF introductory explanation for completeness we are about to move on to the actual linguistic python dictionaries sections.