search

leichtrhino / ChimeraNet

Unofficial implementation of music separation model by Luo et.al.
MIT License
13 stars 2 forks source link

folder format #1

Closed HIN0209 closed 5 years ago

HIN0209 commented 5 years ago

Hello! It is easy to use and seems to work. Before pursuing more with my own data, please guide me on the data placement. I placed the wav. files as shown below. Is this the correct way to process melody1 and vocal1, etc as pairs?? Adding more readme would be appreciated. Thanks!!

root- | |--melody-- |-melody1.wav |-melody2.wav |--vocal
|-vocal1.wav |-vocal2.wav

leichtrhino commented 5 years ago

Thanks to ask a question! The program should run on that folder structure. Actually, a melody file and a vocal file do not have to be a pair. The train program chooses a audio file from each directory, and it loads desired length from chosen file, and it makes mixture data on the fly.

I will enhance readme with a example soon.

HIN0209 commented 5 years ago

@arity-r Thanks for commenting. Can you clarify further? What you say means that the wav files in the respective folders are mixed in a random way (e.g., vocal1 & melody5, instead of vocal1-melody1, vocal2-melody2), correct? My dataset (environmental noise, etc) would be fine with the random combination, but I am wondering if mixing voice and melody from different songs may make sense.

leichtrhino commented 5 years ago

@HIN0209

What you say means that the wav files in the respective folders are mixed in a random way (e.g., vocal1 & melody5, instead of vocal1-melody1, vocal2-melody2), correct?

Correct!

I am wondering if mixing voice and melody from different songs may make sense.

My aim was to extract speaking voice instead of singing voice. So I did not think about this matter.

HIN0209 commented 5 years ago

Thank you again. Let me pursue with my dataset and see how it goes.