source_separation_mirror / config_mdx23c_drumsep2025.yaml
xavriley's picture
Upload config_mdx23c_drumsep2025.yaml with huggingface_hub
54937fb verified
audio:
chunk_size: 523776
dim_f: 1024
dim_t: 1024
hop_length: 512
n_fft: 2048
num_channels: 2
sample_rate: 44100
min_mean_abs: 0.000
model:
act: gelu
bottleneck_factor: 4
growth: 128
norm: InstanceNorm
num_blocks_per_scale: 2
num_channels: 128
num_scales: 5
num_subbands: 4
scale:
- 2
- 2
training:
batch_size: 2
gradient_accumulation_steps: 4
grad_clip: 0
instruments:
- kick
- snare
- toms
- hh
- cymbals
lr: 1.0
patience: 200
reduce_factor: 0.95
target_instrument: null
num_epochs: 1000
num_steps: 1533
q: 0.95
coarse_loss_clip: true
ema_momentum: 0.999
optimizer: prodigy
other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental
use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
loss_multistft:
fft_sizes:
- 2048
hop_sizes:
- 1024
win_lengths:
- 2048
window: "hann_window"
sample_rate: 44100
perceptual_weighting: false
scale: "mel"
n_bins: 128
w_sc: 0.0
w_log_mag: 1.0
w_lin_mag: 0.0
w_phs: 0.0
mag_distance: "L1"
inference:
extension: flac
batch_size: 2
dim_t: 512
num_overlap: 4
normalize: false