Speech Recognition: What can a person with 'electronics' background do to excel in areas such as AI and NLP, which are mostly dominated by computer science principles?

Take an AI class (according to your profile, you are still in school right?), followed by going to graduate school (preferably in the U.S.) to learn about NLP and other topics of interest. Update: Most colleges and universities outside the U.S. do not have rigorous advanced classes. See my EECS blog: https://eecs_ece-and-cs.quora.com for more information (e.g., https://eecs_ece-and-cs.quora.com/Choosing-a-Graduate-Program-in-VLSI-Design-Related-Areas-Things-to-Consider). In advanced EECS classes in U.S. research universities, you will get a good mix of practical experience and theoretical knowledge. You may have a good mentoring and training system at a company to help you learn NLP in the industry. However, they probably will not care about helping you learn the theoretical background.

There are things that are better handled as part of an integrated circuit than in software. The reason for that is mostly performance related as the processing doesn't need to exit in memory through buses. Some areas in AI/NLP are limited by the bottleneck of performance and could not reach real world deployment use-cases because of that reality. I think a good means to step into AI/NLP is to find these cases and work with a team in software to provide an integrated circuit solution path that delivers improvement in performances. One of the cases I had was related to syntactic analysis where I had to perform permutations (look at the syntactic analysis section in here: http://www.codeproject.com/Articles/51026/True-Natural-Language-Understanding-through-a-Conc). The big-O notation for the time required to perform the analysis is O(N^3) where N is the amount of words/POS provided to it. This is a mathematical fact and cannot be avoided. Yet, if the same processing would reside in an integrated circuit, there would be a linear decrease in the amount of time required and would most probably deliver a better outcome. I'd say a good way to enter the field of AI/NLP would be to develop an integrated circuit to perform conceptual analysis also. This is a long dream of mine to produce a thinking computer, but for that to happen, an integrated circuit highly optimized to work with concepts (as described in http://www.codeproject.com/Articles/33306/The-Building-of-a-Knowledge-Base-Using-C-and-an-In and http://www.codeproject.com/Articles/51026/True-Natural-Language-Understanding-through-a-Conc) would need to be produced. Today's processors are specialized to perform mathematical calculations and they do a good job at it. I think we need a paradigm-shift in processors and get them to think, and one good methodology would be through predicate calculus as I documented. To put that methodology into an integrated circuit and then invent a new language that takes full advantage of the processing (C/C++ are not adequate for that) could open up computers to a new golden age through a true thinking automated processor. Good luck!