Ethanfel
5f29b225b7
ComfyUI nodes for UniverSR (ICASSP 2026) — vocoder-free audio super-resolution (8/12/16/24 kHz → 48 kHz) via flow matching. - UniverSR Model Loader: presets auto-download to models/universr, plus local dir / raw .pth (from_local) loading, with caching. - UniverSR Super-Resolution: chunked overlap-add for long audio, per-channel stereo, seed control with global-RNG isolation, wet/dry blend, and an optional before/after spectrogram. - Vendors the universr inference package under vendor/ (prefers an installed copy); only extra dep beyond ComfyUI's stack is torchdiffeq. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
87 lines
1.4 KiB
YAML
87 lines
1.4 KiB
YAML
seed: 42
|
|
|
|
wandb:
|
|
project_name: "UniverSR"
|
|
entity: null # set to your wandb username or team
|
|
run_name: "audio"
|
|
notes: ""
|
|
|
|
dataloader:
|
|
batch_size: 4
|
|
num_workers: 4
|
|
prefetch_factor: 2
|
|
persistent_workers: True
|
|
pin_memory: True
|
|
|
|
collator:
|
|
sampling_rates_probs:
|
|
8: 0.7
|
|
12: 0.1
|
|
16: 0.1
|
|
24: 0.1
|
|
validation_probs:
|
|
8: 1.0
|
|
|
|
dataset:
|
|
common:
|
|
num_samples: 32767
|
|
sr: 48000
|
|
train:
|
|
file_list: "./data/train.txt"
|
|
val:
|
|
file_list: "./data/val.txt"
|
|
|
|
path:
|
|
class_path: universr.flow.path.OriginalCFMPath
|
|
init_args:
|
|
sigma_min: 1.0e-4
|
|
|
|
transform:
|
|
window_fn: 'hann'
|
|
n_fft: 1024
|
|
sampling_rate: 48000
|
|
hop_length: 512
|
|
alpha: 0.2
|
|
beta: 1
|
|
comp_eps: 1.0e-4
|
|
|
|
model:
|
|
in_channels: 2
|
|
out_channels: 2
|
|
dims: [96, 192, 384, 768]
|
|
depths: [2, 2, 4, 2]
|
|
drop_path: 0
|
|
time_dim: 256
|
|
cond_dim: 384
|
|
total_freq_bins: 512
|
|
hr_freq_bins: 432
|
|
feature_enc_layers: 4
|
|
cond_dropout_prob: 0.1
|
|
sr_to_lr_bins: {8: 80, 12: 128, 16: 170, 24: 256}
|
|
|
|
scheduler:
|
|
type: CosineLR
|
|
init_args:
|
|
num_warmup_steps: 10000
|
|
num_training_steps: 5000000
|
|
|
|
optimizer:
|
|
lr: 2.0e-4
|
|
betas: [0.9, 0.99]
|
|
|
|
train:
|
|
num_epochs: 200
|
|
max_steps: 5000000
|
|
ckpt_save_dir: ./ckpts/audio/
|
|
ckpt_load_path: null
|
|
log_step_interval: 1000
|
|
val_step_interval: 50000
|
|
num_val_log_samples: 5
|
|
val_ode_steps: 4
|
|
val_max_sec: 5
|
|
|
|
eval:
|
|
ode_steps: 4
|
|
guidance_scale: 1.5
|
|
max_batches: null
|
|
num_log_samples: 6 |