philippherzig commited on
Commit
740429a
·
verified ·
1 Parent(s): 2f9a174

Move from philippherzig to gridshiftstudio org

Browse files
Files changed (3) hide show
  1. README.md +39 -0
  2. drumsep.onnx +3 -0
  3. drumsep_manifest.json +50 -0
README.md ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ tags:
4
+ - audio
5
+ - source-separation
6
+ - drumsep
7
+ - onnx
8
+ - onnxruntime
9
+ library_name: onnx
10
+ ---
11
+
12
+ # DrumSep (ONNX) — Gridshift mirror
13
+
14
+ Drum-element separation model (kick / snare / cymbals / toms) in ONNX format,
15
+ moved from `philippherzig/drumsep-onnx` to
16
+ [Gridshift Studio](https://huggingface.co/gridshiftstudio) org for
17
+ release-pipeline consistency.
18
+
19
+ Weights and format are unchanged from the previous repository. The original
20
+ upstream is [inagoy/drumsep](https://github.com/inagoy/drumsep),
21
+ a DemucsHT fine-tune on drum stems.
22
+
23
+ ## Usage
24
+
25
+ Consumed by [Gridshift](https://gridshift.studio)'s drum-element separation
26
+ feature via the Rust stem-splitter crate. The `drumsep_manifest.json` in this
27
+ repo lists the ONNX model file with SHA-256 for integrity verification.
28
+
29
+ - **Audio format:** 44.1 kHz stereo, 1 764 000 samples per inference window
30
+ - **Stems:** kick, snare, cymbals, toms
31
+ - **Format:** ONNX (opset 14)
32
+ - **File size:** ~335 MB
33
+
34
+ ## License and attribution
35
+
36
+ Inherited MIT license from the upstream `inagoy/drumsep` project. The ONNX
37
+ conversion was done from the original HDemucs checkpoint using
38
+ `torch.onnx.export` (see `app/rust/drum-separation/convert_drumsep_to_onnx.py`
39
+ in the Gridshift source tree).
drumsep.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ecb8509383ccd437d84e12c216b002a00a2d4e88601e29f2711ea581f207c92a
3
+ size 335071223
drumsep_manifest.json ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "drumsep_v1",
3
+ "version": "1.0.0",
4
+ "backend": "onnxruntime",
5
+ "format": "onnx",
6
+ "opset": 14,
7
+ "sample_rate": 44100,
8
+ "window": 1764000,
9
+ "hop": 882000,
10
+ "stems": ["kick", "snare", "cymbals", "toms"],
11
+ "input_layout": "planar",
12
+ "output_layout": "planar",
13
+ "inputs": [
14
+ {
15
+ "name": "mix",
16
+ "layout": "[1, 2, T]",
17
+ "dtype": "float32",
18
+ "shape": ["1", "2", "1764000"]
19
+ },
20
+ {
21
+ "name": "mag",
22
+ "layout": "[1, 4, F, Frames]",
23
+ "dtype": "float32",
24
+ "shape": ["1", "4", "2048", "1723"]
25
+ }
26
+ ],
27
+ "outputs": [
28
+ {
29
+ "name": "freq_out",
30
+ "layout": "[1, 4, 4, F, Frames]",
31
+ "dtype": "float32",
32
+ "shape": ["1", "4", "4", "2048", "1723"]
33
+ },
34
+ {
35
+ "name": "time_out",
36
+ "layout": "[1, 4, 2, T]",
37
+ "dtype": "float32",
38
+ "shape": ["1", "4", "2", "1764000"]
39
+ }
40
+ ],
41
+ "artifacts": [
42
+ {
43
+ "file": "drumsep.onnx",
44
+ "sha256": "ecb8509383ccd437d84e12c216b002a00a2d4e88601e29f2711ea581f207c92a",
45
+ "size_bytes": 335071223,
46
+ "url": "https://huggingface.co/philippherzig/drumsep-onnx/resolve/main/drumsep.onnx"
47
+ }
48
+ ],
49
+ "entry": "drumsep.onnx"
50
+ }