Friday 27 August 2010

Speech recognition in Linux

Whenever I'm looking at specimens under the microscope and noticing differences, I find it very difficult to stop what I'm doing, look at a bit of paper, and write it down. I'd much prefer to talk about it while looking at the specimen.

The first solution is to record yourself while talking. Audacity is a free, open-source music editing program that is pretty decent. I don't know how useful hard-core sound engineers would find it, but it's not bad for the application that I'm wanting to use it for, namely, recording my voice while I waffle on about what a specimen looks like. I could then listen to the recording repeatedly and transcribe what I say. More efficiently though, I'd be keen for some sorta sound recognition software a la Dragon NaturallySpeaking. NaturallySpeaking is the biggest-selling voice recognition software, and by all accounts it's pretty good. It does require some coin though, so I'm looking for less expensive, preferably open-source programs.

Unfortunately, it doesn't seem like there's much out there. A wikipedia page is a good entry point to the problem. Apparently one of the biggest issues is the lack of a voice database to test algorithms on. To solve this issue, VoxForge has been set up to encourage people to upload recordings and to work on the problem. The Ubuntu Wiki also has a page giving a bit of a road map of what Ubuntu want to see done. It looks like it might be a good project to start working on.

No comments: