EURASIP Journal on Audio, Speech, and Music Processing

Table 2 Various hyperparameters chosen for WaveGAN

From: Transformer-based ensemble method for multiple predominant instruments recognition in polyphonic music

Name	Value
WavGAN Latent dimension	100
Number of channels	1
WavGAN dimension	32
Training batch size	64
Kernel length	25
Generation length	65,536 samples
Loss	WGAN-GP (λ =10)
D updates per G updates	5

Back to article page