Timeline for How to compare 2 matrices of MFCCs, keeping the time information
Current License: CC BY-SA 3.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Dec 20, 2013 at 17:52 | history | edited | pichenettes | CC BY-SA 3.0 |
edited body
|
| Dec 20, 2013 at 16:23 | comment | added | sebpiq | And about computing the average of pair-wise distances, I thought about it, but instead of zero-padding, I thought about interpolating the mfccs frames to have the same amount in the 2 matrices to compare. On the other hand, I can't believe there isn't a better tool than euclidian distance to do this. | |
| Dec 20, 2013 at 16:10 | comment | added | sebpiq | Hi! Thanks for the answer. I actually used an implementation of DTW that doesn't have any option (mlpy.sourceforge.net/docs/3.5/dtw.html). The thing is with DTW, I might have DTW("Pum pum tchak", "Puuum pum tchak") > DTW("Pum pum tchak", "Puuum puuum tchaaak"). While for me, the most important infos are 1) the rhythm, 2) the timbre. Unfortunately, I can't transcribe automatically those audio samples. There's too much of them. | |
| Dec 20, 2013 at 14:05 | history | edited | pichenettes | CC BY-SA 3.0 |
added 271 characters in body
|
| Dec 20, 2013 at 13:59 | history | answered | pichenettes | CC BY-SA 3.0 |