Evaluating algorithms in real world examples is an important step to verify those work in real challenging situations. We recorded an amateur singer to compose a 6-voice montage which permits to evaluate in real world examples but having ground-truth sources to compare with.

To do so, we have chosen The Circle of Life song. The reasons are:
  • Two-languages song, english and zulu.
  • Humming-like sounds which are more challenging than separating lyrics.
  • Synchronized voices and chorals which are slightly delayed on time.
We also add an instrumental backtrack in order to provide musical coherence and sonority. 
Song: The Circle of Life    
Compositor: Elton John  
Owner: Walt Disney Studio 
Duration: 4:34  
Voices (top left, clockwise): Backtrack, Antelope 2, Lioness, Lion, Rafiki, Antelope 1.

The original recordings can be found in GDrive

Encoder: h264
Pixel format: yuv 420p
Framerate: 30 FPS
Montage shape: 2000x3000

Audio encoder: wav pcm_s32be 
Samplerate: 44.1 KHz
2 Channels

Ready-to-use files can be found in GDrive

Framerate: 25 FPS  
Video format: npy (numpy arrays)
Colorspace: RGB  
Array shape: 6850x128x96x3 (TxHxWxC)
Sample Rate: 16384 Hz  
Audio format: wav pcm_s16le mono