SCNet, ONNX (int8 weights)

SCNet (Tong, Zhu, Chen, Kang, Jiang, Li, Wu, Meng, "SCNet: Sparse Compression Network for Music Source Separation", ICASSP 2024, arXiv:2401.13276): band-split convolutions around dual-path LSTMs on the complex spectrogram, 10.1 M parameters (SCNet-large at half the width), trained by its author on MUSDB18-HQ. It splits a song into four stems: drums, bass, other, vocals. This is the network between its STFT and iSTFT, as @audio/neural-separate runs it (model: 'scnet').

File	Size	SHA-256
scnet.int8.onnx	12.9 MB (12,900,277 bytes)	98228931494151762a1c4ab1ec7899a894b1f81fd4a509921a9d2f9bacc50845

The float32 export it is made from is 42.8 MB; float16 weights would be 23.2 MB.

Source

• Model and code: starrytong/SCNet (MIT), at 5d95bf96b19c3eede63248d171efeca8e3abb948.
• Checkpoint: scnet_checkpoint_musdb18.ckpt (SHA-256 1bc0d1abb20bfdf966dcd07637bafd03e4bc13653d09ef18bc9b3e342eafe2aa), release v.1.0.6 of ZFTurbo/Music-Source-Separation-Training (MIT), with its config_musdb18_scnet.yaml; the model code that repository's models/scnet at 84b1eac0887756b4f1a9d7a1ff49105939749ed2.

Licence and attribution

MIT, Copyright (c) 2024 starrytong (LICENSE). The weights' author, in starrytong/SCNet#35 (2026-07-17):

I confirm that the released SCNet and SCNet-large pretrained weights are distributed under the MIT License, consistent with the source code. You are welcome to redistribute the original checkpoints and format-converted versions, including ONNX exports, as part of your MIT-licensed tool, with appropriate attribution.

SCNet by its authors (starrytong/SCNet); the checkpoint as Music-Source-Separation-Training (Roman Solovyev) distributes it; ONNX export and compaction by audiojs. Trained on MUSDB18-HQ (Rafii et al., 2019), licensed for educational use; whether that reaches the weights no project has settled.

@inproceedings{tong2024scnet,
  title     = {SCNet: Sparse Compression Network for Music Source Separation},
  author    = {Tong, Weinan and Zhu, Jiaxu and Chen, Jun and Kang, Shiyin and Jiang, Tao and Li, Yang and Wu, Zhiyong and Meng, Helen},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  year      = {2024},
  eprint    = {2401.13276},
  archivePrefix = {arXiv}
}

Graph

One 11 s segment (485,100 samples at 44.1 kHz, padded to 476 frames) per run:

	name	shape
input	mix_spec	[1, 4, 2049, 476]	STFT: n 4096, hop 1024, no window, scaled by 1/√4096, centered with reflect padding; L re, L im, R re, R im
output	stems_spec	[1, 16, 2049, 476]	each source (drums, bass, other, vocals), channel, re and im

The segments (every 2.75 s), their fades and the input's normalization follow Music-Source-Separation-Training's demix(); @audio/neural-separate's README, Algorithm, has them.

import separate from '@audio/neural-separate'
let { stems } = await separate([left, right], { sampleRate: 44100, model: 'scnet' })

Export, compaction, verification

scripts/export-scnet.py --model scnet --verify exports the network (its rFFT over time as cosine and sine products, its GroupNorm statistics reduced axis by axis) and compares the graph with SCNet.forward on noise and tones: max |diff| ≤ 2.2e-6 of max |y|; the package's pipeline matches SCNet.forward on its segments to 123–133 dB SNR per stem. scripts/compact.py --model scnet --calibrate <two MUSDB18 training previews> makes this file from it:

• Weights: 77 of 82 stored in int8 (99.2 % of the values; symmetric, a scale per output channel, an LSTM's per gate row and direction), the five layers ending the decoder in float16 (decoder.2.0's convolution, decoder.1.1's three transposed convolutions, decoder.2.1's first): rounded alone to int8, each moves the output 27 to 39 dB under its power; all 82 together, 23.4 dB.
• Computed in float32 on every backend: each weight is Cast and multiplied by its scale in the graph, which onnxruntime folds at load (the session holds float32 weights).
• Folded and named short, changing no value: with the input's shape fixed, every value computable from the weights and the shapes alone is stored as the graph computes it (its DFT matrices, made in float64 from a Range, which onnxruntime-web's WebGPU session cannot place); node and value names are base-36 counters.
• Against the export, on the calibration previews: max |diff| 1.1e-2 of max |y|, SNR 42.4 dB.

Quality

The 50 MUSDB18 test previews, BSSEval v4 SDR (museval), the median over songs, dB:

	vocals	drums	bass	other
export (float32)	9.88	9.43	8.35	6.15
this file	9.88	9.44	8.35	6.14
change per song: median · the song that lost most	+0.00 · −0.29	−0.00 · −0.04	−0.00 · −0.10	−0.01 · −0.07

The −0.29 dB is a song whose vocal stem is near silence (PR - Happy Daze, −2.2 dB SDR as exported); with all 82 weights in int8 (12.8 MB) it lost 2.6 dB, hence the five in float16. Remixes (the input plus (g − 1) times a stem, against the true remix): vocals +6 dB 18.23 → 18.23, vocals −6 dB 21.05 → 21.05, drums −6 dB 20.91 → 20.92. Float16 weights (23.2 MB) change no median by more than 0.001 dB.