Core Concepts
Developing realistic digital models of dynamic range compressors using deep learning and state-space models.
Stats
"There are 87 540 s training data."
"Models are trained using the SignalTrain dataset training split with batch size 32 in 60 epochs."
"The testing audio data are segmented with length 223 (≈190.218 s at 44.1 kHz) to test the model’s long-term generalizability."
Quotes
"Virtual analog modeling concerns the digital simulation of analog audio devices like synthesizers and audio effect units."
"Our approach is based on the structured state space sequence model (S4), as implementing the state-space model has proven to be efficient at learning long-range dependencies."
"The need for a model with greater objective accuracy and perceptual quality that is causal, parameter efficient, and real-time capable remains."