AEmotionStudio commited on
Commit
291b781
·
verified ·
1 Parent(s): fec9778

Add LavaSR v2 bundled mirror (enhancer + UL-UNAS denoiser)

Browse files
README.md ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # LavaSR — bundled mirror (v2)
2
+
3
+ Self-contained inference bundle for the MAESTRO desktop app. One ~55 MB download → fully offline 48 kHz speech enhancement.
4
+
5
+ ## Contents
6
+
7
+ - `enhancer_v2/model.safetensors` — Vocos-based bandwidth-extension weights (converted from upstream `pytorch_model.bin` via `safetensors.torch.save_model` for shared-buffer safety)
8
+ - `enhancer_v2/config.yaml` — Vocos hyperparameters (verbatim from upstream)
9
+ - `denoiser/denoiser.safetensors` — UL-UNAS denoiser weights (converted from upstream `denoiser.bin`)
10
+
11
+ ## Source
12
+
13
+ - Enhancer + denoiser bundle: [YatharthS/LavaSR](https://huggingface.co/YatharthS/LavaSR)
14
+ - LavaSR code: [github.com/ysharma3501/LavaSR](https://github.com/ysharma3501/LavaSR) — Vocos backbone adapted for speech BWE, ~5000× realtime on GPU.
15
+ - UL-UNAS denoiser: [github.com/Xiaobin-Rong/ul-unas](https://github.com/Xiaobin-Rong/ul-unas).
16
+
17
+ ## License
18
+
19
+ - LavaSR (enhancer): Apache-2.0
20
+ - UL-UNAS (denoiser): MIT (Xiaobin-Rong)
21
+
22
+ Both upstreams permit commercial redistribution with attribution.
denoiser/denoiser.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dfa7bdff7198b4d80860b943c78f1419f68726bbe637cd35fa16af2252118a1d
3
+ size 733720
enhancer_v2/config.yaml ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ feature_extractor:
2
+ class_path: vocos.feature_extractors.MelSpectrogramFeatures
3
+ init_args:
4
+ sample_rate: 44100
5
+ n_fft: 2048
6
+ hop_length: 512
7
+ n_mels: 80
8
+ padding: same
9
+ f_min: 0
10
+ f_max: 8000
11
+ norm: "slaney"
12
+ mel_scale: "slaney"
13
+
14
+ backbone:
15
+ class_path: vocos.models.VocosBackbone
16
+ init_args:
17
+ input_channels: 80
18
+ dim: 512
19
+ intermediate_dim: 1536
20
+ num_layers: 8
21
+
22
+ head:
23
+ class_path: vocos.heads.ISTFTHead
24
+ init_args:
25
+ dim: 512
26
+ n_fft: 2048
27
+ hop_length: 512
28
+ padding: same
enhancer_v2/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b2be25ec0773ccdb34a4b2196babc557eba762357ff6138dde7ce2861f12e79
3
+ size 56293752