From: Frequency-dependent auto-pooling function for weakly supervised sound event detection
Fourier setting | Training parameters setting | ||
---|---|---|---|
Sampling rate | 32 kHz | Batch size | 18 |
Window size | 2048 | Iterations | 20,000 |
Overlap | 1024 | Optimizer | ADAM |
Mel bands | 64 | Learning rate | 1e −3, 1000 iter*0.9 |