AudioLabs - Two Selected Examples

Two Selected Examples

Retrieval Results

Metadata				Ranks
ComposerID	WorkID	PerformanceID	BM_ThemeID	$\mathcal{C}_{\mathrm{IIR}}^{}$	$\mathcal{C}_{\mathrm{MEL}}^{}$	$\mathcal{C}_{\mathrm{SFM}}^{}$	$\mathcal{C}_{\mathrm{BG1}}^{}$	$\mathcal{C}_{\mathrm{DNN1}}^{}$	$\mathcal{C}_{\mathrm{DNN2}}^{}$	$\mathcal{C}_{\mathrm{CNN}}^{}$	$\mathcal{C}_{\mathrm{MEL}}^{\mathrm{\ast}}$	$\mathcal{C}_{\mathrm{BG1}}^{\mathrm{\ast}}$	$\mathcal{C}_{\mathrm{CNN}}^{\mathrm{\ast}}$
Bach	BWV0848-02	Belder	B306	1	2	1	1	1	1	1	30	1	1
Bach	BWV0884-01	Belder	B378	286	660	1	1	53	4	1	100	128	1

Explanation

The music corresponding to query B306 is monophonic and the melody extraction method $\mathcal{C}_{\mathrm{BG1}}^{\mathrm{\ast}}$ successfully tracks the fundamental frequency of the theme. As a result, this representation achieves a low separation indicator of $\rho=0.195$ . Rank 1 is also achieved by omitting the melody extraction step ( $\mathcal{C}_{\mathrm{BG1}}$ ), but the separation indicator is not as good ( $\rho=0.452$ ).

In contrast, the music corresponding to query B378 is polyphonic and the melody extraction step fails to track the fundamental frequency of the theme. Accordingly, $\mathcal{C}_{\mathrm{BG1}}^{\mathrm{\ast}}$ achieves a bad rank of 128 ( $\rho=1.615$ ). Omitting the melody extraction step ( $\mathcal{C}_{\mathrm{BG1}}$ ) leads to rank 1 for this query ( $\rho=0.760$ ).

Audio Examples

BM_ThemeID	Sonification of Query	Recording	Pianist	License & Link to Recording
B306			Kimiko Ishizaka	Creative Commons Attribution 4.0
B378			Marco Alejandro Gil Esteva	Creative Commons Attribution 3.0

Note that the presented recordings are only for illustrative purposes and are not the same as the audio recordings in the data set BM-Medium, which are used for the experiments.

Sheet Music

B306 (Theme)
B306 (Full Score)
B378 (Theme)
B378 (Full Score)

back

International Audio Laboratories Erlangen

Two Selected Examples

Retrieval Results

Explanation

Audio Examples

Sheet Music