============== njoliat / notes / mostly from Aug 3 2014 ================================ step 1: do some crappy prototyping by finding the "hottest" parts of [the hottest?] songs and concatenating them. for hottness look at density of soundcloud comments or something. https://developers.soundcloud.com/docs/api/reference https://developers.soundcloud.com/docs https://developers.soundcloud.com/docs/api/sdks https://github.com/soundcloud/soundcloud-python todos normal: getting some kind of 'recent hot tracks' list (e.g. from https://soundcloud.com/explore) current issue is soundcloud.com/explore is not a user, it's some other kind of weird stream thing. * maybe i can get this via API somehow? * can't get via api? try screen scraping? [done, would be nice to scrape infinite-scrolling thing] getting raw audio data, and then associating comment timestamps with moments in that data simple density metric on comment timestamps to find 'hottest part' of song stitching together the hottest parts of multiple audio tracks weird: doing sentiment analysis on comments?1 superimposing text-to-speech comments over the audio, using all the different text to speech voices? possible ways to get / combine the hottest section(s) of a song? * take the N milliseconds with the most comments * take a small section of audio after EVERY comment; concatenate sections. could even, like, layer sections * can do these things with raw milliseconds or could do it cleanly with beats/bars, e.g. by using echonest. could even take the independent / non-contiguous hottest 5 beats or whatever? if we wanted to go that route, could even change speeds / beatmatch stuff. that runs the risk of being listenable / boring, though.
git clone git://git.numm.org/beats-capital