| Well I think CMUSphinx and Julius have actually made leaps and bounds. Jasper is a thin layer built on top of those systems (sphinx in this case). I guess I'm just easy to impress - I think the fact that we have off-the-shelf open source software that does voice recognition with any kind of quality is amazing. And again, a lot of the time, it's about how well your model is trained. One of the ideas I had was to work training into the product, and have the user say sentences over time (maybe ~5 a week, at random times) to help the machine understand better (something like "hey, if you have a minute, let's work on our communication"). Also, even if the voice is a little bit computer-y, I don't mind, yeah you won't match on feature set right out of the gate, but I don't mind. Also, there's an opportunity for advancement in that area -- if I were to build a site that collected voice/utterance samples from voice artists and had users pay a little bit for packs, I think we could fund the open-source teams doing the work to make more realistic voice/utterance. I think of Jasper as a first step, but then again, I am biased, I like the idea and have been wanting to do it myself for a long time |