System.Speech and Neural Networks

Question

I am not sure where to post this. However, it's a general question. I program voice activated software, and I had a thought.

If I was to program a speech neural network. What options would be best to take?

I know that Aforge has machine learning with Fuzzy Logic. However, if I wanted to start from scratch.

I would be using back propagation, and possibly recursive learning. Would there be a way I could extract the default speech sound wave files that MS Speech uses, instead of recording every single word in the English language into a sound wave.

I have programmed feed forward and back propagation neural networks before.

My question is at this moment, is there a way to pull the sound wave files to implement the inputs, or will I need to record each word into a sound wave?

My long term goal is within the next 3 years is to have my program distinguish sentence structure, then the program would be able to piece together the nouns, verbs, adverbs, etc., etc.,. So it can create its own sentences to ask the user a question.

I don not want to use open source except for the sound waves. I can handle the coding. I just need to know if I can pull MS Speech sound waves or do I need to record them myself?

Answer 1

Natural Language Processing (NLP) is a Text Based approach to Machine Comprehension. There are some very good papers that may give you a place to start:

https://openreview.net/pdf?id=B1-q5Pqxl

and

https://arxiv.org/pdf/1611.09830v2.pdf

A MATCH-LSTM Neural Network Architecture seems to be the current state of the Art for as you say:

My long term goal is within the next 3 years is to have my program distinguish sentence structure, then the program would be able to piece together the nouns, verbs, adverbs, etc., etc.,. So it can create its own sentences to ask the user a question.

Some example code was published by Wang and Jiang : https://github.com/shuohangwang/SeqMatchSeq

Your goal is board, but I believe achievable. Massive milestones have been reached already. Good Luck!