Training Music Sample Identification Models on Real Sample Pairs

R. Oguz Araz; Joan Serrà; Xavier Lizarraga-Seijas; Xavier Serra; Yuki Mitsufuji; Dmitry Bogdanov

cs.SDSep 18, 2026

Training Music Sample Identification Models on Real Sample Pairs

Authors: R. Oguz Araz, Joan Serrà, Xavier Lizarraga-Seijas, Xavier Serra, Yuki Mitsufuji, Dmitry Bogdanov

Organizations: Music Technology Group, Universitat Pompeu Fabra, Spain · Sony Europe · Sony CTC America

Abstract

Sample identification (SI) is the task of matching pairs of tracks, where one track is created by musically transforming an element of the other. In the absence of sample annotations at scale, the dominant training paradigm has depended on artificially creating sample pairs. Although a recently released dataset provides annotations of real sample pairs at scale, an effective training recipe is missing. In this work, we present SI Embeddings (SIE), an SI model that achieves state-of-the-art results on three benchmarks, including a large-scale test set. We show that the previous state of the art trained on artificial pairs generalizes only partially to real pairs, and that its training data limits its performance. We also show that real pairs do not fully account for SIE's performance: its architecture and training recipe contribute substantially. We provide the first fully supervised training recipe for real-world SI, establishing a strong foundation for future research in the field.

Figures & tables

System	Training Pairs	Dim.	Sample100		SamplePairs
System	Training Pairs	Dim.	mAP ( $\uparrow$ )	mNAR ( $\downarrow$ )	mAP ( $\uparrow$ )	mNAR ( $\downarrow$ )
Van Balen et al. [ 16 ]	None (rule-based)	–	0.390 †	n/a	n/a	n/a
Cheston et al. [ 9 ]	Artificial	2048	0.441 †	n/a	n/a	n/a
SampleID [ 13 ]	Artificial	2048	0.603	7.6	0.450	4.7
SampleID-R (ours)	Real	2048	0.651	6.3	0.511	4.1
SIE (ours)	Real	1024	0.758	3.9	0.701	2.4

Table 1: Comparison of SI systems on the Sample100 and SamplePairs benchmarks. Dim. denotes embedding dimensionality. SampleID-R is our retraining of SampleID on real pairs. † denotes scores cited from corresponding publications; n/a denotes scores that are neither published nor computable by us due to lack of available implementation.

Model	mAP ( $\uparrow$ )		mNAR ( $\downarrow$ )
SampleID [ 13 ]	0.226	$\pm$ 0.007	19.4	$\pm$ 0.5
SampleID-R	0.299	$\pm$ 0.008	14.3	$\pm$ 0.4
SIE	0.389	$\pm$ 0.009	13.0	$\pm$ 0.4
NMFP [ 1 ]	0.101	$\pm$ 0.005	35.4	$\pm$ 0.6
CLEWS [ 15 ]	0.181	$\pm$ 0.007	32.4	$\pm$ 0.7
Fish [ 3 ]	0.207	$\pm$ 0.007	29.9	$\pm$ 0.6

Table 2: Comparison of models on the WhoSampled130K test set. SampleID-R is our retraining of SampleID on real pairs. The last three models were not trained for sample identification (see text). Numbers following

\pm

are 95% confidence interval half-widths for the mean over 10,444 queries.

Figure 1: Effect of training SIE on nested random subsets of the WhoSampled130K training set pairs, with everything else held fixed. Evaluated on the WhoSampled130K test set. Dashed horizontal lines mark the scores of SampleID, taken from Table 2 .

SM	AD	mAP ( $\uparrow$ )		mNAR ( $\downarrow$ )
✗	✗	0.320	$\pm$ 0.008	16.2	$\pm$ 0.5
✗	✓	0.334	$\pm$ 0.008	16.1	$\pm$ 0.5
✓	✗	0.365	$\pm$ 0.009	13.0	$\pm$ 0.4
✓	✓	0.389	$\pm$ 0.009	13.0	$\pm$ 0.4

Table 3: Ablation of signal manipulation (SM) and audio degradation (AD) during SIE training, evaluated on the WhoSampled130K test set.

Loss	mAP ( $\uparrow$ )		mNAR ( $\downarrow$ )
Triplet	0.389	$\pm$ 0.009	13.0	$\pm$ 0.4
NT-Xent ( $\tau_{0}=0.01$ )	0.332	$\pm$ 0.008	12.6	$\pm$ 0.4
NT-Xent ( $\tau_{0}=0.05$ )	0.331	$\pm$ 0.008	13.0	$\pm$ 0.4
NT-Xent ( $\tau_{0}=0.10$ )	0.320	$\pm$ 0.008	13.0	$\pm$ 0.4

Table 4: Comparison of triplet and NT-Xent losses for training SIE, evaluated on the WhoSampled130K test set.

\tau_{0}

is the initialization of the learnable temperature.

Segment Duration (s)	mAP ( $\uparrow$ )		mNAR ( $\downarrow$ )
5	0.384	$\pm$ 0.009	12.7	$\pm$ 0.4
6	0.389	$\pm$ 0.009	13.0	$\pm$ 0.4
8	0.381	$\pm$ 0.009	13.1	$\pm$ 0.4

Table 5: Effect of segment duration on SIE, evaluated on the WhoSampled130K test set. Each model uses its training segment duration at inference, with hop fixed at 3 s and batch size fixed across models.

Explore similar work

Building a Dataset for Music Sample Identification
Sep 14, 2026R. Oguz Araz, Xavier Lizarraga, Xavier Serra +1Music Structure Analysis Symbolic Music Generation
Music Technology Group, Universitat Pompeu Fabra, Spain
Music Transcription with (Almost) No Supervision
May 22, 2026Saebyeol Shin, Chao Wan, Zhenzhen Liu +4Multi-Instrument Music Transcription Transcription
Cornell University, Ithaca, NY
Turning music identification into a neural forward pass
Jun 15, 2026Muhammad Taimoor Haseeb, Ahmad Hammoudeh, Gus XiaMusic Structure Analysis Neural Audio
Music X Lab, Mohamed Bin Zayed University of Artificial Intelligence, UAE

Sep 14, 2026, cs.SDJ/K move · Enter open · S save

Building a Dataset for Music Sample Identification

R. Oguz Araz, Xavier Lizarraga, Xavier Serra, Dmitry Bogdanov

Music Technology Group, Universitat Pompeu Fabra, Spain

Music Structure Analysis Symbolic Music Generation

Open paper AskPDF arXiv

Sample identification (SI) is the task of matching an element of a musical work to its musically transformed versions used to create new works. The task has received little attention and lacks large-scale publicly available data. In this work, we mine sampling annotations from a music database and split them for training and evaluation. The resulting dataset is nearly three orders of magnitude larger than the existing SI benchmarks, with training, validation, and test sets of 114 k, 6 k, and 10 k tracks. We find that naively splitting the annotations places the same tracks in different sets. To avoid this, we construct a graph from the annotations and split it over connected components. We further find that a single mega-component contains half of the annotations, making component-wise splitting incompatible with balanced splits; we trim it, yielding a leakage-aware pipeline. We share the dataset for non-commercial scientific research purposes only and make the data-analysis and splitting code publicly available. We hope that our work fosters research on SI.

Abstract

Explore similar work

Building a Dataset for Music Sample Identification

Music Transcription with (Almost) No Supervision

Turning music identification into a neural forward pass