In this page, we introduce a few audio examples drawn from the test set of MUSDB18. These audio examples are compressed and reconstructed using 4 different models: Baseline (DAC trained on MUSDB18 only), RandRVQ1, RandRVQ4 and RandRVQ5 (which are introduced in table II of the paper).
It is important to have in mind that the RandRVQ models are still random at inference. The results displayed here might not be the best results we could get from those models.
Experiments results
Sample 1
Source
Baseline
RandRVQ1
RandRVQ4
RandRVQ5
Sample 2
Source
Baseline
RandRVQ1
RandRVQ4
RandRVQ5
Sample 3
Source
Baseline
RandRVQ1
RandRVQ4
RandRVQ5
Sample 4
Source
Baseline
RandRVQ1
RandRVQ4
RandRVQ5
Sample 5
Source
Baseline
RandRVQ1
RandRVQ4
RandRVQ5
Sample 6
Source
Baseline
RandRVQ1
RandRVQ4
RandRVQ5
Sample 7
Source
Baseline
RandRVQ1
RandRVQ4
RandRVQ5
Sample 8
Source
Baseline
RandRVQ1
RandRVQ4
RandRVQ5
Sample 9
Source
Baseline
RandRVQ1
RandRVQ4
RandRVQ5