Saturday, November 17, 2007

Closed

I see we have a response to our last post. Thank you, highperformance, for showing an interest. For anyone who also stumbles on this blog, I'll just post the final update.

We completed the project. Although the results were far from what we hoped to achieve. Due to the exceedingly large number of neurons, the BLSTM is very slow. Also, after having it run for several days on end, it has yet to show any conclusive evidence of actually positively identifying phonemes.

We started this project as part of the seminary Speech Recognition; we have since handed in our paper and received a passing grade. Both Timo and myself considered investing more time into this project, possibly as a study project. However this did not fit into either of our curricula so far. That makes the project abandonware at this point.

The last version of the source code is available here:
www.haploid.nl/sr2k7/lstm-070620-1511.zip

I am not aware of any restrictions on the use of this code, so use at your leisure. This code comes as is, with very little documentation on its use.