---
title: bs_polarformer
canonical_url: "https://www.modelscope.cn/models/ymp123/bs_polarformer"
md_url: "https://www.modelscope.cn/models/ymp123/bs_polarformer.md"
repository: ymp123/bs_polarformer
last_updated: 2026-08-22
license: mit
base_model:
  - ZFTurbo/Music-Source-Separation-Training
base_model_relation: finetune
library_name:
  - onnx
downloads: 0
stars: 0
tags:
  - music-source-separation
  - vocal-separation
  - onnx
  - webgpu
  - audio
---

# bs_polarformer

> bs_polarformer - ymp123 在 ModelScope 开源的模型。BS PolarFormer – ONNX Vocal Separation

ymp123/bs_polarformer 是 ModelScope 魔搭社区上的机器学习模型，采用 mit 许可，基于 ZFTurbo/Music-Source-Separation-Training 构建。

- **Repository**: ymp123/bs_polarformer
- **License**: mit
- **Base model**: ZFTurbo/Music-Source-Separation-Training
- **Tags**: music-source-separation, vocal-separation, onnx, webgpu, audio
- **Downloads**: 0
- **Stars**: 0
- **Last updated**: 2026-08-22

Source: https://www.modelscope.cn/models/ymp123/bs_polarformer

---

# BS PolarFormer – ONNX Vocal Separation

ONNX conversion of the **BS PolarFormer** vocal separation model from
[Music-Source-Separation-Training](https://github.com/ZFTurbo/Music-Source-Separation-Training).

BS PolarFormer is a BSRoformer architecture with **PoPE** (Polar Positional Embeddings)
instead of rotary embeddings. It separates **vocals** from **other** (instrumental) in stereo audio at 44.1kHz.

## Files

| File | Size | Description |
|------|------|-------------|
| `bs_polarformer.onnx` | 201 MB | FP32 ONNX model (core: band split → transformers → mask estimator) |
| `bs_polarformer_fp16.onnx` | 103 MB | FP16 quantized (weights stored as float16, ~same quality) |
| `model_bs_polarformer_float16.yaml` | 3.6 KB | Model config |
| `convert_to_onnx.py` | 19 KB | Conversion script (PyTorch → ONNX) |
| `run_onnx_inference.py` | 7 KB | CLI inference script |
| `index.html` | 18 KB | Web app (runs in browser via WebGPU/WASM) |

## Architecture

The ONNX model contains only the **core neural network** (51M parameters):

```
Audio → [STFT] → Core Model (ONNX) → [Mask] → [iSTFT] → Vocals
                  ├─ BandSplit (60 frequency bands)
                  ├─ 12× (TimeTransformer + FreqTransformer)
                  │   └─ 8-head attention, dim=256, PoPE embeddings
                  └─ MaskEstimator (2-layer MLP per band)
```

STFT/iSTFT are handled outside the ONNX model (in PyTorch or JavaScript).

**Input:** `(batch, time_frames, 4100)` — interleaved stereo STFT features (1025 freq × 2 channels × 2 real/imag)

**Output:** `(batch, 1, 2050, time_frames, 2)` — complex mask

## Quality (vs PyTorch reference)

| | FP32 ONNX | FP16 ONNX |
|---|---|---|
| Mask max abs diff | ~1e-7 | ~4e-5 |
| Audio SNR | 107 dB | 48.6 dB |
| Pearson correlation | 1.00000000 | 0.99999642 |
| Model size | 201 MB | 103 MB |

Both are perceptually identical to the PyTorch model. The original model achieves **SDR 11.00** on vocals (Multisong Dataset).

## Usage

### Python (ONNX Runtime)

```bash
pip install onnxruntime librosa soundfile pyyaml einops torch

# Download this repo, then:
python run_onnx_inference.py song.mp3 --output_dir output/
python run_onnx_inference.py song.mp3 --fp16  # use smaller model
```

### Browser (WebGPU)

Serve the files with any HTTP server and open `index.html`:

```bash
python -m http.server 8080
# Open http://localhost:8080
```

Drop an audio file, select FP32 or FP16, and click "Separate Vocals". Uses WebGPU when available, falls back to WASM.

### Convert from scratch

```bash
# Download checkpoint
wget https://github.com/ZFTurbo/Music-Source-Separation-Training/releases/download/v1.0.20/model_bs_polarformer_float16.ckpt

# Convert
python convert_to_onnx.py          # FP32 only
python convert_to_onnx.py --fp16   # FP32 + FP16
```

## Credits

- Original model & training: [ZFTurbo/Music-Source-Separation-Training](https://github.com/ZFTurbo/Music-Source-Separation-Training)
- BSRoformer architecture: [lucidrains](https://github.com/lucidrains)
- PoPE embeddings: [PoPE_pytorch](https://pypi.org/project/PoPE-pytorch/)