Hey Kyle, we didn't try anything more advanced than next-step sampling. You probably have a better sense than I do how much improvement such techniques are likely to yield. My unfounded suspicion is that we're close to the limit of generation quality from this dataset, and so I'm most interested in trying to gather 10-100x more skilled performances, one way or another.
There's also no consensus on whether the high- or low-temperature samples sound better. I've heard both opinions from several people.
Sageev did the final rendering, not sure what he used but I'm pretty sure it was nothing too fancy.
By far my favorite use of Hangouts is the crossword puzzle app. It's much better than having 5 people try to collaboratively solve a crossword in person.
I have no idea why I haven't seen more games like this in Hangouts...
Other people being willing to pay for something doesn't necessarily make it positive-sum or economically productive. A lot of startups (certainly not all) seem to be trying to win tournaments: http://en.wikipedia.org/wiki/Tournament_theory
In that sense, SF is is similar to Hollywood. And the result is that society devotes too many resources to the next big social network or the next big summer blockbuster, and not enough to, say, making comfortable chairs.
> If there is a 1% chance the world will end unless we do x, we shouldn’t do a cost-benefit analysis. Instead, assuming x is feasible, we should simply do it.
What is the chance the world will end even if we do x? What does feasible mean? Is killing 10% of the population feasible? Is this potential end of the world happening in a year, or in a thousand years?
I'd go on, but this is starting to look a bit like a cost-benefit analysis.
My favorite thing about this is that he's extremely good at playing it. I also have an instrument app that allows for continuous pitch + autotuning (http://www.iansimon.org/fingrophone/), but his looks like it allows for way more expressivity.
One nontrivial part is transforming the spectrogram into some representation that is robust to the things that can affect the query audio, like background noise. Another nontrivial part is figuring out, given this representation, how to quickly match the query with a song or database of songs.