MDX23C InstVoc HQ as a float16 Core AI asset

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+stems-mdx23c-instvoc-float16.aimodel/main.mlirb filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,65 @@
+---
+license: mit
+pipeline_tag: audio-to-audio
+tags:
+  - source-separation
+  - vocal-separation
+  - mdx23c
+  - core-ai
+  - apple-silicon
+---
+
+# MDX23C InstVoc HQ — Core AI (float16)
+
+Vocal / instrumental separation for Apple silicon: **MDX23C-8KFFT-InstVoc_HQ**
+converted to a Core AI asset (`stems-mdx23c-instvoc-float16.aimodel`, 214 MB) for
+on-device use from Swift.
+
+## Credit
+
+- **Architecture:** TFC-TDF-UNet v3 (MDX23C) — Kim & Lee.
+- **Implementation:** [ZFTurbo/Music-Source-Separation-Training](https://github.com/ZFTurbo/Music-Source-Separation-Training) (MIT).
+- **Weights:** `MDX23C-8KFFT-InstVoc_HQ` from [Ultimate Vocal Remover](https://github.com/Anjok07/ultimatevocalremovergui)'s public model repository (MIT).
+
+This repository only converts those weights; all credit for the model belongs to them.
+
+## What the asset computes
+
+The network between the spectrogram and the stem spectrograms — the STFT, chunking and
+inverse STFT run in the host:
+
+- input `spec` `[1, 4, 4096, 256]` float32 — left re, left im, right re, right im; bins 0–4095
+  of an STFT with n_fft 8192, hop 1024, periodic Hann, centred; one chunk is 261,120 samples
+  at 44.1 kHz.
+- output `stems` `[1, 8, 4096, 256]` float32 — vocals (4 channels), then instrumental.
+
+Weights and arithmetic are float16 behind float32 input and output.
+
+## Measured (macOS 27, Apple silicon)
+
+| | this asset | float32 export | 8-bit Core ML package |
+|---|---|---|---|
+| one chunk vs PyTorch float32 | 62.6 dB | 123.9 dB | 35.3 dB |
+| time per chunk (GPU) | 0.27 s | 0.31 s | 0.74 s |
+| a 30 s mix, end to end | 3.0 s | 3.5 s | 6.9 s |
+
+On three 30-second mixes with a known vocal, all three scored the same vocal and
+instrumental SDR to within 0.01 dB. Repeated runs are bit-for-bit identical. The export
+asserts the converted network equals upstream's `model(chunk)` exactly before saving.
+
+## Use
+
+With [`stems`](https://github.com/arraypress), which fetches this on first use:
+
+```sh
+stems song.mp3        # song-vocal.wav + song-instrumental.wav
+```
+
+From Swift, with [swift-vocal-isolation](https://github.com/arraypress/swift-vocal-isolation):
+
+```swift
+let separator = try await MDXSeparator(contentsOf: assetURL)
+let stems = try await separator.separate(contentsOf: songURL)
+```
+
+Converted with `Tools/export_mdx.py --precision float16` in swift-vocal-isolation.

stems-mdx23c-instvoc-float16.aimodel/main.hash

@@ -0,0 +1 @@
+(R��"�����Z�T���c?�N�P��;A��

stems-mdx23c-instvoc-float16.aimodel/main.mlirb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2852bce622b386cec1c85acf165491088bf48a633fff4e8850bec13b41f505fc
+size 224073232

stems-mdx23c-instvoc-float16.aimodel/metadata.json

@@ -0,0 +1,8 @@
+{
+  "creationDate" : "20261010T143633Z",
+  "author" : "Kim & Lee — TFC-TDF-UNet v3 (MDX23C); implementation ZFTurbo (MIT); weights UVR MDX23C-8KFFT-InstVoc_HQ (MIT); Core AI export by VocalIsolation",
+  "producer" : "coreai-core 1.0.0b2",
+  "license" : "MIT",
+  "description" : "MDX23C vocals\/instrumental: spec [1,4,4096,256] (L re, L im, R re, R im; n_fft 8192, hop 1024, 261,120 samples) → stems [1,8,4096,256] (vocals ×4, instrumental ×4). STFT, chunking and iSTFT in the host.",
+  "assetVersion" : "2.0"
+}