| Name | Size | Uploaded | Xet hash |
|---|---|---|---|
| LICENSE | 1.08 kB xet | 525f89c6 | |
| README.md | 1.01 kB xet | 23d6ecf0 | |
| accompaniment.onnx | 39.3 MB xet | 71137e1b | |
| vocals.onnx | 39.3 MB xet | 878c3b33 |
Vocal remover (browser build)
Unmodified Spleeter 2-stem ONNX files from csukuangfj/sherpa-onnx-spleeter-2stems revision 7001ba316a615cacddb3f9ef3ec416661a277e26:
vocals.onnx— 39,318,336 bytesaccompaniment.onnx— 39,318,343 bytes
The weights are Deezer's pretrained Spleeter 2stems model (deezer/spleeter, MIT, see LICENSE), converted to ONNX by the sherpa-onnx project (Apache-2.0 code). This is the fast vocal/instrumental pair WaveSpeed documents for the vocal remover.
Input x float32 [2, splits, 512, 1024]: magnitude STFT of a stereo 44,100 Hz clip (n_fft 4096, hop 1024, Hann window, center false, first 1024 bins, frames padded to a multiple of 512). Output y is the same shape. The two files are combined as a soft mask. The file metadata field sample_rate=41000 is an upstream typo; separation uses 44,100 Hz, matching sherpa-onnx scripts/spleeter/separate_onnx.py.
- Total size
- 493 MB
- Files
- 34
- Last updated
- Nov 8
- Pre-warmed CDN
- US EU US EU