User Tools
hpss_for_beat_tracking
====== Differences ====== This shows you the differences between two versions of the page.
|
hpss_for_beat_tracking [2013/06/30 16:53] phunter created |
hpss_for_beat_tracking [2013/06/30 17:22] (current) phunter |
||
|---|---|---|---|
| Line 1: | Line 1: | ||
| - | ====== Example Project Title ====== | + | ====== Source Separation for Beat Tracking ====== |
| | Authors | Hunter McCurry | | | Authors | Hunter McCurry | | ||
| Line 8: | Line 8: | ||
| ===== - HPSS ===== | ===== - HPSS ===== | ||
| - | A section | + | I was looking for a simple and fast Harmonic / Percussive source separation method that could be implemented as my an analysis pre-processing step. The idea is to take the spectrogram and filter out high-frequency variation in each the X (time) and Y (frequency) dimension to achieve the harmonic and percussive components, respectively. |
| - | ==== - LibROS Implementation ==== | + | ==== - Median Filtering ==== |
| - | You can cite something like this((some paper title, author, etc.)) | + | Derry FitzGerald has a paper called Harmonic/Percussive Source Separation Using Median Filtering that uses Median Filtering (a relatively straight-forward image processing technique) to do this.((http://dafx10.iem.at/proceedings/papers/DerryFitzGerald_DAFx10_P15.pdf)) |
| - | ===== - Beat Tracking ===== | + | ==== - LibROSA Implementation ==== |
| - | A section | + | Brian McFee wrote an implementation of this algorithm and included it in his LibROSA library.((https://github.com/bmcfee/librosa)) |
| - | ==== - Evaluation code ==== | + | ==== - [iPython demo] ==== |
| - | You can cite something like this((some paper title, author, etc.)) | + | ==== - [Audio examples] ==== |
| - | ===== - Results ===== | + | ===== - Beat Tracking ===== |
| + | I used a dataset of 102 audio files (30-second samples) for which I had human-tagged truth values in the form of beat locations. I ran the Echo Nest analyzer (version 3.2.0) on the unaltered audio files and subsequently on the percussive component of my HPSS. | ||
| + | ==== - Evaluation code ==== | ||
| + | Turns out it is not easy to evaluate beat-tracking algorithms with a single, simple metric. There are many proposed algorithms for doing this. In the paper Evaluation Methods for Musical Audio Beat Tracking Algorithms((Matthew E. P. Davies, Norberto Degara Aand Mark D. Plumbley)) is a comparison of 10 different methods of comparison. | ||
| + | ===== - Results ===== | ||
hpss_for_beat_tracking.1372625610.txt.gz ยท Last modified: 2013/06/30 16:53 by phunter
Page Tools
Except where otherwise noted, content on this wiki is licensed under the following license: CC Attribution-Share Alike 3.0 Unported
