Skip to content
Audio

bs-roformer-sw-onnx-vocals

A vocal ONNX extractor with quantization

2 min read

This model is a specialized version of the BS-RoFormer SW-Fixed vocal channel architecture (44.1 kHz stereo), rewritten for onnxruntime to run on a CPU. Unnecessary tools were removed from the graph, making the file up to 12 times smaller than the original and speeding up inference. The model outputs a clean vocal track and is released in several precision variants, including quantized int8 versions. This tool will be useful for musicians, sound engineers, and audio application developers looking for high-quality track separation.

Key features

  • Reduced size: from 58 to 211 MB depending on the variant (3.4 to 12.2 times smaller than the 6-stem version).
  • CPU support: runs via onnxruntime without requiring a dedicated graphics card.
  • Various formats: fp32, int8wo (weights-only), and dynamic int8 variants are available.
  • High quality: the model successfully passes rigorous correlation tests (corr > 0.999) compared to the base GGUF CLI.

What you can do

  • Extract clean vocals from a musical composition using a Python script or CLI utility.
  • Integrate a vocal splitter into your own local audio processing pipelines.
  • Compare speed and accuracy between int8dyn_pc (for fast inference) and int8wo_b32 (for maximum fidelity to the original).

Example

import soundfile as sf
from bsroformer_sw_onnx import Splitter

splitter = Splitter(
    repo_id="remy-st/bs-roformer-sw-onnx-vocals",
    filename="bs_roformer_sw6_vocals_int8dyn_pc.onnx",
    num_stems=1,
    stem_names=("vocals",),
)
mix, sr = sf.read("song.wav")
vocals = splitter.separate(mix)["vocals"]
sf.write("vocals.wav", vocals, sr)
Read moreDetails
• Model weights: from 58.2 MB (int8dyn) to 211 MB (fp32).
• Metrics: int8wo_b32 shows a Pearson correlation of 0.9999892 on long audio clips.
• Requirements: onnxruntime, numpy, soundfile, CPU execution.
• Limitations: original weight licensing is officially undefined; recommended for non-commercial or private use.
Discuss Telegram 1

Will it run?

Runs in FP32

Fits without quantisation.

  • FP32weights 0.2 GB · ready files0.7 GB
  • BF16weights 0.1 GB0.6 GB
  • FP8 / INT8recommendedweights 0.0 GB0.6 GB
  • 4-bitweights 0.0 GB0.5 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 · FP32 — 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.

    # koidn/bs-roformer-sw-onnx-vocals
    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 koidn/bs-roformer-sw-onnx-vocals --local-dir ~/models/bs-roformer-sw-onnx-vocals
    # The model is in ~/models. How to run it is in the model card (link above).
  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 BS-ROFO-SW-Fixed (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

koidn/bs-roformer-sw-onnx-vocals
Author
koidn
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