nexiwave's CUDA based Speech Indexing solution is featured in NVIDIA's latest CUDA Review. To recap, with a total rewrite of acoustic scoring, we moved more than 75% of computation into CUDA 3.0. Our further tuning achieved more than 75 times speedup.
Since the release of CUDA based solution into our production environment, we have been thinking really hard on both technical front, as well as on marketing directions. On the technical side, we are quite interested in moving the remaining computing cycles into CUDA hardware. However, as I explained in my other post, we realized that speed is just one side of the benefit. The very nature of massive computing power, as well as massive memory available, has made certain carefully crafted algorithm obsolete. More things are certainly cooking in nexiwave. More news to come on this front soon.
The whole CUDA Review can be found here:
http://www.nvidia.com/content/newsletters/web/CUDA_Week_in_Review_June_18_10.html