Embedded Files
Current Openings
Septro-Temporal Decoupling in CNNs:
We present a different way of answering the question of what does a network learns by decoupling the information learned by the CNN in spectral and temporal domain.
Applications to:
Understanding raw waveform based CNNs for automatic speech recognition
Exploiting the concept of separable filters to design LRC (Low Rank Convolutions) based CNNs.
Generalizing CNNs with Rank-k convolutions for footprint reduction.
Hierarchical multistream (spectral and temporal ) speech processing using CNNs.
End-to-End Speaker identification and verification using raw waveform based CNNs
r-vector system: A robust speaker identification system; an embedding system for speaker verification.
r-rector with adaptation: An end-to-end speaker identification system
Acoustic scenes classification
Learning hierarchy aware embedding from raw audio for ASC
Page updated
Google Sites
Report abuse