Sample types v1: CLAP + trained head, 39 types (68.3% held out)
Browse files- .gitattributes +3 -0
- README.md +35 -0
- clap-audio.aimodel/main.hash +1 -0
- clap-audio.aimodel/main.mlirb +3 -0
- clap-audio.aimodel/metadata.json +8 -0
- head-b1.f32 +0 -0
- head-b2.f32 +0 -0
- head-labels.txt +39 -0
- head-w1.f32 +3 -0
- head-w2.f32 +0 -0
- mel-filters.f32 +3 -0
.gitattributes
CHANGED
|
@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
clap-audio.aimodel/main.mlirb filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
head-w1.f32 filter=lfs diff=lfs merge=lfs -text
|
| 38 |
+
mel-filters.f32 filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
pipeline_tag: audio-classification
|
| 4 |
+
tags:
|
| 5 |
+
- audio-classification
|
| 6 |
+
- music
|
| 7 |
+
- samples
|
| 8 |
+
- clap
|
| 9 |
+
- core-ai
|
| 10 |
+
- apple-silicon
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# muse sample types
|
| 14 |
+
|
| 15 |
+
Names what kind of sound a sample is, from **39 types** — kick, snare, clap, hat_closed,
|
| 16 |
+
bass_loop, sub_bass, pluck, pad, riser, vocal_hit and more (`head-labels.txt`).
|
| 17 |
+
[`muse`](https://github.com/arraypress) uses it for its `class` column.
|
| 18 |
+
|
| 19 |
+
**68.3% right on held-out sample packs** (packs not seen in training; chance is
|
| 20 |
+
about 3%). Most misses are between types a person would also confuse, such as
|
| 21 |
+
sub_bass against bass_hit.
|
| 22 |
+
|
| 23 |
+
## How it works
|
| 24 |
+
|
| 25 |
+
1. `mel-filters.f32` — a 64-band mel spectrogram (48 kHz, 1,001 frames).
|
| 26 |
+
2. `clap-audio.aimodel` — the CLAP audio encoder,
|
| 27 |
+
[laion/clap-htsat-unfused](https://huggingface.co/laion/clap-htsat-unfused)
|
| 28 |
+
(LAION, Apache-2.0), converted to Core AI with its weights unchanged; it turns
|
| 29 |
+
the spectrogram into a 512-number embedding.
|
| 30 |
+
3. `head-*.f32` — a small trained head on the L2-normalised embedding:
|
| 31 |
+
512 → 512 (GELU) → 39, softmax. Little-endian float32: `w1` [512×512], `b1`
|
| 32 |
+
[512], `w2` [512×39], `b2` [39].
|
| 33 |
+
|
| 34 |
+
Only the head was trained, on a licensed sample library; the audio is not
|
| 35 |
+
distributed and cannot be recovered from these files. Apache-2.0, following CLAP.
|
clap-audio.aimodel/main.hash
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
��}��w���UBgdhAO4�jJ���
|
clap-audio.aimodel/main.mlirb
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bcb27d80f07790ba835542676468414f34da6a4afbbb950df80ecf5c67392546
|
| 3 |
+
size 132854940
|
clap-audio.aimodel/metadata.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"producer" : "coreai-core 1.0.0b2",
|
| 3 |
+
"creationDate" : "20260827T182520Z",
|
| 4 |
+
"license" : "Apache-2.0",
|
| 5 |
+
"description" : "CLAP audio encoder (laion\/clap-htsat-unfused) converted to Core AI. Mel spectrogram [B,1,1001,64] in, 512-d projected joint-space audio embedding out. Weights unchanged; format conversion only.",
|
| 6 |
+
"assetVersion" : "2.0",
|
| 7 |
+
"author" : "LAION"
|
| 8 |
+
}
|
head-b1.f32
ADDED
|
Binary file (2.05 kB). View file
|
|
|
head-b2.f32
ADDED
|
Binary file (156 Bytes). View file
|
|
|
head-labels.txt
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
acid
|
| 2 |
+
arp
|
| 3 |
+
atmo
|
| 4 |
+
bass
|
| 5 |
+
bass_hit
|
| 6 |
+
bass_loop
|
| 7 |
+
chord
|
| 8 |
+
clap
|
| 9 |
+
crash
|
| 10 |
+
cymbal
|
| 11 |
+
downer
|
| 12 |
+
drum_fill
|
| 13 |
+
drum_loop
|
| 14 |
+
fx
|
| 15 |
+
hat_closed
|
| 16 |
+
hat_open
|
| 17 |
+
impact
|
| 18 |
+
keys
|
| 19 |
+
kick
|
| 20 |
+
lead
|
| 21 |
+
noise
|
| 22 |
+
pad
|
| 23 |
+
percussion
|
| 24 |
+
piano
|
| 25 |
+
pluck
|
| 26 |
+
reese_bass
|
| 27 |
+
ride
|
| 28 |
+
riser
|
| 29 |
+
shaker
|
| 30 |
+
snare
|
| 31 |
+
stab
|
| 32 |
+
strings
|
| 33 |
+
sub_bass
|
| 34 |
+
sweep
|
| 35 |
+
synth_hit
|
| 36 |
+
synth_loop
|
| 37 |
+
tom
|
| 38 |
+
vocal_hit
|
| 39 |
+
vocal_loop
|
head-w1.f32
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8619987d18be7f5b3f1ff565c5f68eb02bb814c3220b93c7571a17c1400003a9
|
| 3 |
+
size 1048576
|
head-w2.f32
ADDED
|
Binary file (79.9 kB). View file
|
|
|
mel-filters.f32
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d643673b1bd3ff90fca67ebfbd4341246c2faed2a8a1adb2cc8b589a5b514c1d
|
| 3 |
+
size 131328
|