Local AM/FM modulations estimation: application to audio sinusoidal modeling and blind source separation


Audio signal analysis/synthesis results



The results are expressed in terms of RQF (measured in dB between reference analyzed signal and the reconstruction through the estimated parameters).
Sound name Method
SM08SM12t2w2w6
speech8.087.527.828.216.89
singing voice14.4014.0215.0515.1313.62
saxophone28.5627.8927.1529.9023.71
drums6.546.526.726.634.29

Audio blind source separation results



The results are expressed in terms of RQF, SIR, SDR, SAR computed through BssEval

Title: AlexanderRoss_VelvetCurtain


mix - ref source 1 - ref source 2
MethodResultsRQFSIRSDRSAR
(w2)
Oracles1 s2 8.70/9.6619.02/23.218.08/9.248.50/9.44
CFM + kmeanss1 s2 5.88/6.029.77/11.144.58/4.846.58/6.32
CAM + kmeanss1 s2 2.47/2.6211.70/2.570.23/1.490.84/9.96
CFM/CAM + kmeanss1 s2 2.99/3.1612.16/3.650.633/1.931.20/8.36
(t2)
Oracles1 s2 8.53/9.4518.32/23.947.91/9.018.39/9.17
CFM + kmeanss1 s2 5.84/5.849.58/9.654.55/4.566.65/6.62
CAM + kmeanss1 s2 2.42/2.4312.48/2.53-0.10/1.360.38/9.54
CFM/CAM + kmeanss1 s2 2.30/2.3011.78/2.44-0.72/1.17-0.19/9.08
(SM12)
Oracles1 s2 9.05/9.6019.14/23.588.50/9.188.94/9.36
CFM + kmeanss1 s2 5.43/5.388.57/8.633.97/3.926.38/6.27
CAM + kmeanss1 s2 2.48/2.5711.19/2.53-0.14/1.330.50/9.41
CFM/CAM + kmeanss1 s2 2.38/2.4711.53/2.46-0.45/1.270.12/9.44