Projects
Adaptive time-frequency learning
A research file connecting learned wavelets, wavelet packets, denoising and fractional superlets.
active
Multiresolution decompositions describe a signal at several scales. The linked studies make parts of that representation trainable and combine them with neural models for speech emotion recognition or enhancement.
SigWavNet and wavelet-packet models explore learned filters and thresholding. NWPA investigates a bidirectional autoencoder for enhancement. LFST learns fractional superlet representations with a spectro-temporal encoder. These are related methods, not interchangeable implementations.
Research directions
- 01
Research programs
Learnable signal representations
Wavelets, wavelet packets and fractional superlets as adaptive representations for emotion analysis and speech enhancement.