espeak, espeak-ng, pyttsx3, and MBROLA

Posted by: @codecage

Do you any insight into if you trained Pocket Sphinx new words using the microphone where you get 100% understanding if it then can do a better job of recognizing that same word now spoken with the microphone not in that optimal position?

This topic is not easy to address in this post.  In fact, this is going to be a more advanced topic when we actually get into Pocket Sphinx.

The "Training" is not what you are imagining.   And, to be perfectly honest, I'm not clear on exactly how training works in Pocket Sphinx yet as I haven't yet studied that feature in detail.  

The training is done using a program called SphinxTrain.   And, I'm not even sure how that program actually works yet as I haven't yet used it.   However, I do have some basic understanding "which could be wrong!"

I don't yet know whether SphinxTrain can be used to add new words to the dictionary.  My current guess is that it doesn't do that, unless it allows you to manually add a word by first typing the word in?  I'm just not there yet.

My current, very limited, understanding of SphinxTrain, is that it is used to improve the phoneme recognition of words that are already in the dictionary.    Or at least words that you have already typed into a text file.   Whether it actually adds new words from these text files to the dictionary on its own I don't yet know.

I don't think it's possible to have SpinxTrain create new words in the dictionary just by you speaking into the microphone.  After all, if that were the case it would need to take a very wild guess at how they should be spelled.  So I'm guessing that it doesn't even work that way at all.

What I believe it is used for is to train PocketSphinx to better recognize words that are already in the dictionary, or at least in a training text file that you have already typed in.

I'm very anxious to look into all of that.  However, there were so many other things to learn just to get Pocket Sphinx up to the point where I currently have it, including the Python interface, which is something that needs to be added, because you could use Pocket Sphinx with any programming language, so there are many other ways to access it.

~~~~

On a far more distant note I will say that following:

I'm planning on having my robot be able to learn new words via Pocket Sphinx SRE, by speaking the word into the computer and having the robot ask me how to spell the word, etc. 

However, all of that will be part of my own Linguistic AI program that really has nothing to do with Pocket Sphinx other than the fact that my program is using Pocket Sphinx as the SRE tool.

The point being that there are basically infinite possibilities using Pocket Sphinx as a basic SRE tool.  But some of those possibilities will need to come from the programmer's own imagination, creativity and programming.

Hopefully, someday, I'll actually get to a point where I can start making videos on my Linguistic AI project.   This will be a project of my own programming using Python, Numpy, and various AI programming techniques.

At that point eSpeak and Pocket Sphinx will basically just be the fundamental TTS and SRE tools.  By my Linguistic AI program will be the conductor of the main Linguistic orchestration. 😊 

That's where I'd really like to be right now.  But I had to get these TTS and SRE tools under my belt first.

And back to your question.  I will eventually cover Sphinx Train at some point.  However, because I'll be creating my own Linguistic AI program to train my robot in general, I'm not sure exactly how much I'll need to employ Sphinx Train.  That is yet to be discovered.

@codecage

Now  you're going to get me side-tracked into learning about Sphinx Train instead of making the videos on eSpeak.  🤣 

This is always a major problem.  I want to do too many things!   And today I'm working on tractors again!  Trying to clean and salvage old spark plugs because I'm too lazy to go out and buy new ones.   20 miles to town!  I'd rather just clean up the old plugs.