Radioland

I just finished a great conversation with Pam Hill Kroyer on her show Pam Without Boundaries on KFAI. Thanks again to Pam for having me on.

Pam made a somewhat mysterious reference to a show I’m playing this Sunday at Room Zero. She didn’t offer details because the location is a little hush-hush, but you can get the location, pass code, and secret handshake by emailing shield.your.eyes[at]live.com. Also, all three of the tracks Pam played are available at iTunes and for the Steve Jobs averse, Amazon.

During our talk, I alluded to some still vague plans to incorporate speech recognition software into my setup. This idea is still in the research phase, and to be honest, I’m not sure it’s even possible. Commercial speech recognition software can perform poorly in the relative quiet of a home office. It’s uncertain that even the most advance algorithms would handle the noisy atmosphere of a live stage show. But I’m going to give it a try. The recognition API I’ve been looking at is Sphninx 4, developed at Carnegie Mellon University. Since Java is what I know best, I’ve been looking at Java APIs, but I’d welcome suggestions for other APIs. I also have the impression that Hidden Markov Models, which is what Spninx 4 uses, are slightly dated, and the state of the art uses Conditional Random Fields. I was wondering if anyone has a perspective on the relative merits of these models.