---
license: apache-2.0
base_model: RSNuts/Universal_Source_Separation
tags:
  - mlx
  - audio
  - audio-to-audio
  - source-separation
---

# USS ResUNet30 (MLX)

Universal source separation — extracts a chosen sound class from a mixture.
Converted to MLX from the [ByteDance USS](https://github.com/bytedance/uss) code
and the published
[RSNuts/Universal_Source_Separation](https://huggingface.co/RSNuts/Universal_Source_Separation)
checkpoint.

## Files

| file | size |
| --- | ---: |
| `resunet30_fp32.safetensors` | 102 MiB |
| `resunet30_fp16.safetensors` | 51 MiB |

## Precision

| precision | size | speed | peak memory | quality vs fp32 |
| --- | ---: | ---: | ---: | ---: |
| fp32 | 102 MiB | **27.2x** | **1203 MiB** | reference |
| fp16 | **51 MiB** | 25.8x | 1259 MiB | 57.3 dB |

Bold marks the best value in each column. Quality is fp16 against this model's own
fp32 output — the cost of the precision change, not a score for the model.

fp16 halves the download. On the machine below it was also slightly slower and
used slightly more memory than fp32, since the forward pass dominates the
footprint rather than the weights.

Measured on an Apple M1 Pro (16 GB), 30 s of audio, three timed runs per
configuration; speed is realtime factor. These are single-machine figures — expect
different absolute numbers on other hardware, and treat the ordering as indicative
rather than guaranteed.

## Conditioning

USS is conditioned on a 527-dimensional vector over the
[AudioSet](https://research.google.com/audioset/) classes, which selects what the
separator extracts. It is an input rather than a model weight and is not shipped
here: each vector is a normalised multi-hot indicator over a class group — zero
everywhere, `1/n` at each of the `n` chosen indices — so a class list fully
determines it.

Class indices follow AudioSet's `class_labels_indices.csv`, published by Google
with the dataset under CC BY 4.0.

## Used by

[AudioToolSwift](https://github.com/starkdmi/AudioToolSwift)