Please use it because it is in here. Since it uses GPU, the processing speed is quite fast. Google's best to lend you a GPU for free! !!
If you get an error, restart the runtime and re-execute the cells one by one from the preparation chapter. That should solve most of the errors. If you get a network error when downloading, re-execute the cell. It seems to occur depending on the timing of file generation. If you change your browser to Chrome, you may not get an error.
First, open this note in playground mode. It is at the top of the screen. Then execute all the cells in the "Preparation" chapter (ctrl + enter). Then execute the cell (ctrl + enter) in the chapter "Loading Modules and Methods".
You can also execute it by clicking [] in each cell as shown in the picture below.
When you execute the cell in "Upload the song you want to divide from here", the following display will be displayed. Click Browse and upload any sound source. __ Please re-execute this cell every time even if you upload multiple sound sources. __ </ font> When the image below is displayed, the file upload is complete.
Also, when uploading multiple sound sources at the same time, it is also recommended to click the "Upload button" in the folder displayed in "File" on the left tab, or drag and drop to the tab to upload the file. is.
__ When you add a song, please read the cell of viewing etc. again to reflect the song in viewing etc. __ </ font>
To watch, execute the cell of the watching chapter. After a while, the following will be displayed and you can watch it.
You can download it as a zip or one by one.
Please use this block to make a sound source without drums or a sound source with vocals off.
There is a problem that Download and viewing are sometimes executed twice. It will be cured if you restart the runtime. However, since the root cause is unknown and there is no practical problem, it is left unattended. If anyone knows the cause, please let me know. I think it's happening probably because the scope isn't handled properly. I want to study and fix it.
I would like to use Spleeter for the time being, but if you are not good at Google Colab or Python, I recommend using splitter.ai. A very easy-to-understand interface is implemented. However, it seems that the division work is done on the server side, so waiting time is required.
Excerpted from the About section of the Github page, one mixed sound source using deep learning (so-called music that we listen to on a daily basis) ) Is a library that divides into up to 5 parts of vocals, drums, piano, bass, and other musical instruments.
I feel that it will be useful for people who copy by ear and practice of copy bands. I didn't want to knead the command line, so I wrote it in Google colaboratory.
Recommended Posts