| audio: | |
| chunk_size: 261120 | |
| dim_f: 4096 | |
| dim_t: 256 | |
| hop_length: 1024 | |
| min_mean_abs: 0.01 | |
| n_fft: 8192 | |
| num_channels: 2 | |
| sample_rate: 44100 | |
| model: | |
| act: gelu | |
| bottleneck_factor: 4 | |
| growth: 64 | |
| norm: InstanceNorm | |
| num_blocks_per_scale: 2 | |
| num_channels: 64 | |
| num_scales: 5 | |
| num_subbands: 4 | |
| scale: | |
| - 2 | |
| - 2 | |
| training: | |
| batch_size: 6 | |
| coarse_loss_clip: false | |
| datasets: | |
| - ../data/moises/bleeding | |
| ema_momentum: 0.999 | |
| grad_clip: null | |
| instruments: | |
| - Vocals | |
| - Drums | |
| - Bass | |
| - Other | |
| lr: 0.0001 | |
| num_steps: 150000 | |
| q: 0.93 | |
| target_instrument: null | |
| inference: | |
| batch_size: 2 | |
| dim_t: 1024 | |
| num_overlap: 8 |