Skip to content
Audio

Splitwave models

separate songs into vocals and instrumentals right in your browser

2 min readupdated

A collection of optimized ONNX models for splitting audio tracks into vocals, instrumentals, lead vocals, and backing vocals locally on the user's machine via WebGPU or WebAssembly. The repository includes four proven models from UVR, viperx, and aufr33: heavyweight, high-accuracy BS-RoFormer and Mel-Band RoFormer in fp16, alongside fast UVR-MDX-NET models in their original fp32 precision. The main difference from the original checkpoints is adapting the RoFormer architecture to work within WebGPU limits without losing computational accuracy, allowing them to run without a heavy backend. This release is useful for web audio app developers, karaoke services, and musicians.

Key features

  • WebGPU and WebAssembly ready: frequency band splitting in RoFormer is restructured to fit browser rendering limits while preserving the original math.
  • Two modes for different tasks: RoFormer models deliver maximum separation quality, while lightweight MDX-Net models prioritize fast processing.
  • Mixed precision (fp16/fp32): RoFormer weights are converted to fp16 for a smaller footprint, while normalization layers and positional tables remain in fp32 to maintain quality.
  • Integrity verification: includes a manifest.json file with file sizes and SHA-256 hashes to verify downloaded weights inside your app.

What you can do

  • Embed into a web service: launch an online tool for creating backing tracks or extracting acapellas where all processing runs client-side with zero server GPU costs.
  • Build browser-based karaoke: chain vocal separation and lead-vocal removal models (mel_roformer_karaoke_fp16.onnx or mdx_kara_2.onnx) to isolate backing vocals only.
  • Offer quality presets: let users toggle between a fast preview using MDX-Net and a high-quality final render with BS-RoFormer.

Example

The RoFormer models take an 8-second stereo STFT spectrogram chunk of shape [1, 2050, 801, 2] (with n_fft 2048 and hop 441) and output a complex mask, which is multiplied by the original spectrum before running the inverse STFT in your application.

Read moreDetails
Separation quality: BS-RoFormer (ep 317) achieves an SDR of 12.98, and Mel-Band RoFormer karaoke scores an SDR of 10.20.
Included models: bs_roformer_fp16.onnx and mel_roformer_karaoke_fp16.onnx (fp16), mdx_voc_ft.onnx and mdx_kara_2.onnx (fp32).
Licensing: distributed under the original authors' terms (Ultimate Vocal Remover team, viperx, and aufr33).
Discuss Telegram 1

Will it run?

Runs in FP16

Fits without quantisation.

  • FP16weights 0.4 GB · ready files1.0 GB
  • FP8 / INT8recommendedweights 0.2 GB0.7 GB
  • 4-bitweights 0.1 GB0.6 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · FP16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # TheDailyOWO/splitwave-models
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download TheDailyOWO/splitwave-models --local-dir ~/models/splitwave-models
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to splitwave-models
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of splitwave-models (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

TheDailyOWO/splitwave-models
Author
TheDailyOWO
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments