xavriley commited on
Commit
b4c89a3
·
verified ·
1 Parent(s): 7b5eefa

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +23 -0
README.md CHANGED
@@ -1,3 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
1
  # GAPS Dataset (Guitar-Aligned Performance Scores)
2
 
3
  - [Metadata (Google Sheets)](https://docs.google.com/spreadsheets/d/1vxhLMUoAFNPZSRsakqBATSs9f2BDJS0Jb3dbA65v5p0/edit?usp=sharing)
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ tags:
4
+ - guitar
5
+ - mir
6
+ - midi
7
+ - transcription
8
+ - musicxml
9
+ size_categories:
10
+ - n<1K
11
+ ---
12
  # GAPS Dataset (Guitar-Aligned Performance Scores)
13
 
14
  - [Metadata (Google Sheets)](https://docs.google.com/spreadsheets/d/1vxhLMUoAFNPZSRsakqBATSs9f2BDJS0Jb3dbA65v5p0/edit?usp=sharing)
15
+ - [Paper (ArXiv)](https://arxiv.org/abs/2408.08653)
16
+
17
+ 300 solo guitar performances with accurately aligned MIDI and musicxml scores.
18
+
19
+ ## Update: version 1.1
20
+
21
+ - audio now included
22
+ - fixed scorehash case sensitive clashes (thanks Daniel Wägner for pointing this out)
23
+
24
+ ## Abstract
25
+
26
+ > We introduce GAPS (Guitar-Aligned Performance Scores), a new dataset of classical guitar performances, and a benchmark guitar transcription model that achieves state-of-the-art performance on GuitarSet in both supervised and zero-shot settings. GAPS is the largest dataset of real guitar audio, containing 14 hours of freely available audio-score aligned pairs, recorded in diverse conditions by over 200 performers, together with high-resolution note-level MIDI alignments and performance videos. These enable us to train a state-of-the-art model for automatic transcription of solo guitar recordings which can generalise well to real world audio that is unseen during training.