whoa... interesting... did it work out? I mean could you crowdsource voice recognition? In the case of voicebunny words are cheaper on longer texts so you could probably get a better price for reads that are in the thousands of words range. Of course, in here you will pay higher prices for higher quality
I needed to build a audio corpus to train acoustic models on. It actually worked pretty good. I was able to build up about 100 hours worth. It was unbelievably tedious because I had to verify the audio even after I used amazon turk a second time to do a first verification pass.
The other thing is that I don't necessarily want perfect quality audio. I wanted whatever microphone audio normal people have so it probably worked out better that I used amazon turk.
