The above are the chromas based on the acoustic chord transcription and key extraction using Kyogu's paper. The chromas are from a mid file and a sound track of song "Crazy Little Thing Called Love". The chroma form mid file is clearly align with the wav file. Awesome! I see the direction where this project is going. :)
The above Chormas are based on a mid and a sound track of the song "Not in Love". The interesting part is that we can see the pattern somehow align with each other but with a shift. This brings out an important feature of a cover song. Most of the cover song does not follow the original song exactly, for instant, my music band would measure the highest key of a song that I am comfortable with and shift the key of the whole song when we do the cover version. The other point is that we can see the wrap around in the mid file comparing to the wav file, but still captures the pattern of a song.
By the way, the chromas here are extract from full songs, with
Sample rate: 5512.00 Hz
Duration: (the number of chroma of each song + 1)* hop size
Hop size: 1024
FFT frame size: 2048