Upload 2 files
Browse files
SatonoDiamond/diffusion.yaml
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
data:
|
| 2 |
+
block_size: 512
|
| 3 |
+
cnhubertsoft_gate: 10
|
| 4 |
+
duration: 2
|
| 5 |
+
encoder: vec768l12
|
| 6 |
+
encoder_hop_size: 320
|
| 7 |
+
encoder_out_channels: 768
|
| 8 |
+
encoder_sample_rate: 16000
|
| 9 |
+
extensions:
|
| 10 |
+
- wav
|
| 11 |
+
sampling_rate: 44100
|
| 12 |
+
training_files: filelists/train.txt
|
| 13 |
+
unit_interpolate_mode: nearest
|
| 14 |
+
validation_files: filelists/val.txt
|
| 15 |
+
device: cuda
|
| 16 |
+
env:
|
| 17 |
+
expdir: logs/44k/diffusion
|
| 18 |
+
gpu_id: 0
|
| 19 |
+
infer:
|
| 20 |
+
method: dpm-solver++
|
| 21 |
+
speedup: 10
|
| 22 |
+
model:
|
| 23 |
+
k_step_max: 0
|
| 24 |
+
n_chans: 512
|
| 25 |
+
n_hidden: 256
|
| 26 |
+
n_layers: 20
|
| 27 |
+
n_spk: 1
|
| 28 |
+
timesteps: 1000
|
| 29 |
+
type: Diffusion
|
| 30 |
+
use_pitch_aug: true
|
| 31 |
+
spk:
|
| 32 |
+
SatonoDiamond: 0
|
| 33 |
+
train:
|
| 34 |
+
amp_dtype: fp32
|
| 35 |
+
batch_size: 96
|
| 36 |
+
cache_all_data: true
|
| 37 |
+
cache_device: cpu
|
| 38 |
+
cache_fp16: true
|
| 39 |
+
decay_step: 100000
|
| 40 |
+
epochs: 100000
|
| 41 |
+
gamma: 0.5
|
| 42 |
+
interval_force_save: 5000
|
| 43 |
+
interval_log: 10
|
| 44 |
+
interval_val: 2000
|
| 45 |
+
lr: 0.0001
|
| 46 |
+
num_workers: 4
|
| 47 |
+
save_opt: false
|
| 48 |
+
weight_decay: 0
|
| 49 |
+
vocoder:
|
| 50 |
+
ckpt: pretrain/nsf_hifigan/model
|
| 51 |
+
type: nsf-hifigan
|
SatonoDiamond/model_500000.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:743b4145211694c9a80355f87a0d5d28f7213ed3a2467db10185ba9e605b9e74
|
| 3 |
+
size 220893960
|