Staticfree Blog

I got frustrated with the lack of a Lojban text to speech voice for Festival last night, so I put one together. It took a bit of research, but I've grown to like Festival's engine. It's all lisp with a C++ core. As Lojban has an entirely phonetic textual representation, it was easy to hack together a phoneme mapping an existing diphone collection for Lojban. (in other words, I didn't have to record myself speaking to make it go).

In stopping by the Lojban IRC channel, I found out that someone has recorded some diphones for Lojban (albeit with a French accent). Best yet, the recordings are indexed by phoneme-pair. With a bit of work (listening to 1186 wav files and labeling where the key parts of the waveform are) we can have a Lojban text-to-speech that's got its own voice! Of course, I don't have this much time to spend on such, but who knows. Maybe my 16h of bus riding will prove to be productive.