search

mdeff / fma

FMA: A Dataset For Music Analysis
https://arxiv.org/abs/1612.01840
MIT License
2.23k stars 439 forks source link

What about a much larger collection? #1

Closed skinkie closed 7 years ago

skinkie commented 7 years ago

Hi, as radio station we have a much larger collection of lossless encoded audio (FLAC). Would it be interesting to see the performance on our collection?

mdeff commented 7 years ago

Hi, that sounds indeed very interesting. Not only to see the performance, but more importantly to train and evaluate. The problem with music is to be able to distribute the audio to researchers, because of copyright. Are you able to redistribute your collection ? Otherwise, how do you imagine to provide access ?

BTW which radio station are you talking about ? I was not able to found it.

skinkie commented 7 years ago

So Dutch law would exclude education and scientific research. But if I would offer the set, including computing power. I would like to keep the contribution anonymous at this time :)

5 Onder een voordracht, op- of uitvoering of voorstelling in het openbaar wordt niet begrepen die welke uitsluitend dient tot het onderwijs dat vanwege de overheid of vanwege een rechtspersoon zonder winstoogmerk wordt gegeven, voor zover de voordracht, op- of uitvoering of voorstelling deel uitmaakt van het schoolwerkplan of leerplan voor zover van toepassing, of tot een wetenschappelijk doel.

mdeff commented 7 years ago

Would this be valid worldwide for your entire collection ? I'm not knowledgeable with copyright laws. The thing is I don't want to release it and then for copyright holders to come down on us and force us to remove everything. That would waste everybody's time. The songs hosted by FMA are under a permissive Creative Commons license which allows redistribution, so we are sure.

BTW, how many songs and what kind of songs are we talking about ?