Building Jarvis like application for local languages

614 views Asked by At

The jarvis application that is currently developed, is in English. I want to customize it to use local language. How to develop this kind of app for local languages? what kind of programming languages I must know to proceed to the development? I have tested the english version of the jarvis, it works well for me. How to attach the c# with HTK for the purpose of the development?

2

There are 2 answers

1
Nikolay Shmyrev On BEST ANSWER

How to develop this kind of app for local languages?

You don't need to develop from scratch, take existing software and build on it. For example you can consider https://github.com/jasperproject/jasper-client, it's pretty actively developed.

what kind of programming languages I must know to proceed to the development?

Most NLP libraries are in Python or Java. You also need shell scripting (awk/perl) experience because often models are built with Linux tools.

For speech recognition it's easiest to use CMUSphinx, the tutorial to add your language to CMUSphinx is at http://cmusphinx.sourceforge.net/wiki/tutorialam.

I have tested the english version of the jarvis, it works well for me. How to attach the c# with HTK for the purpose of the development?

There are many ways for interoperability:

1) C# can invoke HTK tools as binaries through Process.Start http://msdn.microsoft.com/en-us/library/system.diagnostics.process.start(v=vs.110).aspx

2) You can build a library from HTK and invoke it with PInvoke through interop framework

3) You can build a TCP or HTTP server with HTK tools and connect to this server from C# application to get speech recognition results.

Overall, you could probably use existing solutions like mentioned above, they have all hard things implemented, you only need to configure your local language.

0
Nilkanth Shet Shirodkar On

I would suggest you to go for HTK or if you have lots of training data then go for kaldi one of the best toolkit for speech recognition for local language which uses deep learning.