DEMO_AUDIOSHRINK Decomposition into tonal and transient parts
This demos shows how to do audio coding and "tonal + transient"
decomposition using group lasso shrinkage of two WMDCT transforms
with different time-frequency resolutions.
The signal is transformed using two orthonormal WMDCT bases.
Then group lasso shrinkage is applied to the two transforms
in order to:
select fixed frequency lines of large WMDCT coefficients on the
wide window WMDCT transform
select fixed time lines of large WMDCT coefficients on the
narrow window WMDCT transform
The corresponding approximated signals are computed with the
corresponding inverse, IWMDCT.
Figure 1: Plots and time-frequency images
The upper plots in the figure show the tonal parts of the signal, the
lower plots show the transients. The TF-plots on the left are the
corresponding wmdct coefficients found by appropriate group lasso
shrinkage
Corresponding reconstructed tonal and transient sounds may be
listened from arrays rec1 and rec2 (sampling rate: 44.1 kHz)
Package: ltfat