|
If a project of speech recognition is just to overcome difficulties of speech variations with always writing progrmas by humans, it's in a wrong direction. The project should include program synthesis. The computer has a core program with input, output and program synthesis and management. It can produce programs it needs and use appropriate programs when it is going to do something, and deletes programs which aren't often used. Then the computer can grow, learn and change characteristics. The speech recognition computer then writes programs and updates parameters when facing new kinds of variations of speech and accumulates experience learning to listen, getting feedback and improves itself to better and better accurate. This is a right direction. A project always doing the improvement program writing jobs with humans can't have significant advancing results.
|