Thursday, April 29, 2004

Was just going through old stuff present on my disk. Came across the project I had done when I was in thrid year of graduation. I saw demo of that code again. I saw it again. It has some special importance, and always feels good whenever I play it. I still remember those days when we were doing that project on Speech Recognition. We were trying vaguely to implement algorithms (DFT and FFT), about which we didn't even knew head or tail at that time. Then we had meeting with Mr. Sampath from concepts, and things got quite simple. We abandoned that idea of DFT and FFT and we resorted to the old time domain method for speech recognition. As usual jaydeep and sameer did most of the understanding and coding. Amit was ill at that time, vishal and i had gone home. When I came back, Jaydeep and sammer announced that after many failures with those DFT and FFT algorithms, we finally were able to recognise two words with pretty high success rate. After that we extended the concept to speech operated elevator. Only drawback at that time was that it's success rate used to decrease as we tryied to increase number of words to recognise, and with many different sounds and pronunciations our database would increase and so did the failure rate. Finally we compromised at identifying only two words, one word (soar) to make that lift move up, and other (reverse) to move it down. The idea to identify was to identify the position of letter "s" from these two words. Then we further extended the concept with the real dummy lift, driven by motors. That didn't happen, because of shortage of time, But we did that graphically on computer simulation of the lift.
Never got bored with that stuff, though it might have been old, or nobody might use that technique to recognise speech now. It's always a nice feeling when you could say that I did that when I was in school, and at that time when others were also trying. Though the implementation wasn't any sophisticated, but the concept was relatively new at that time. And for a college project, it was really nice to attemt quite successfully. In the compitition organised at the inter-college level, we did got second prize. Just after another masterpiece, a remote controlled car by nachiket, ajit, rohit, anupam and sumeet. I still like that third year project more that any other project I have done. Partly, the last year project was almost a sort of bouncer to me. It was successful specially because of sameer and jaydeep, as they did the most part of understanding. At the end, when I understood that project completely, it seemed simple enough to implement for anybody. On the way we disproved somebody (don;t want to disclose names), and identified a new relationship between speech signals. We got an international paper in EURASIP journal for that.

No comments:


Related Posts with Thumbnails