search

keunwoochoi / music-auto_tagging-keras

Music auto-tagging models and trained weights in keras/theano
MIT License
616 stars 142 forks source link

what is the 29.1 file of million song? #5

Closed SeekPoint closed 7 years ago

SeekPoint commented 7 years ago

do you refer the subset of one million songs?

keunwoochoi commented 7 years ago

did I mention 29.1 files somewhere? It is 29.1 second and they are preview clips of MSD.

On 29 Oct 2016, at 00:39, yk_data notifications@github.com wrote:

do you refer the subset of one million songs?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

keunwoochoi commented 7 years ago

Oh, probably you mean those lines from readme.

More information about the files are in this repo

SeekPoint commented 7 years ago

do you means just git clone https://github.com/keunwoochoi/MSD_split_for_tagging under the music-auto_tagging-keras ?

keunwoochoi commented 7 years ago
zumpchke commented 7 years ago

I have a question related to this work. I am trying to figure out how to input audio to a CNN. In this work, are all audio clips the same size? Any tips on how to handle variable length audio (i.e, an entire song would be of differing lengths).

as641651 commented 7 years ago

@keunwoochoi 22k? In the paper, you have mentioned about 214k clips. The repository do not contain the clips. So I suppose, you have downloaded the preview clips, trimmed it to 29s and got the mel-spectogram. Right?

keunwoochoi commented 7 years ago

@VanushVaswani All audio clips are the same size. There could be some heuristics on how to get the 29s segments and how to merge the scores. E.g., random, middle, structure segmentation, ..

keunwoochoi commented 7 years ago

@as641651 Thanks for noting it, you're correct. It's 214k clips. Yes, that's correct.