Before last Thursday, the situation was this: using approach A I was getting 0.3% performance improvement in a speech recongizer’s performance.
Last thursday, I beleived I had a got a 0.7% improvement using aproach B. I was so elated I didnt even cross-check and I shot of emails to everyone in the lab about it. Since both these approaches are in the early part of the recognizer, everyone else’s work had to be redone to come up with the best speech recognizer of the lab. Everyone was looking forward to redoing stuff and getting better performance.
Well, late, really late last night, I found out I had done something really stupid and had screwed up the implementation of approach B and through some freak chance, I was getting better performance. On redoing approach B correctly, I was again getting only 0.3% improvement. To put it mildly, I was crushed. I even started considering leaving research. I also felt like an idiot shooting off(sp?) emails like that.
Well, just now, some one hour back, I tried approach C and I get 0.8% impreovement!! This time I have retraced my steps and so far I still havent seen anything weird. Also, since everyonehad taken an extended weekend off, no one had started from where I had left off.
Hoping this it…
why am I even bothering with this? I dont know…I am excited and want to tell this to someone
Thanks for tuning in…