Esc50 Sep Eval

---
name: esc50-sep-eval
description: Evaluates zero-shot language-queried audio source separation on environmental sound classes. The benchmark tests the model's ability to isolate a target sound from a mixed audio mixture using text labels. Use when the user wants to benchmark on ESC-50, or asks about evaluating this task. Reports SDRi.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2308.05037
  bibtex_key: liu2023separate
  confidence: high
---

# esc50-sep-eval

> Separate Anything You Describe — Liu et al. (2023) (arXiv:2308.05037, 2023)

## What this evaluates

Evaluates zero-shot language-queried audio source separation on environmental sound classes. The benchmark tests the model's ability to isolate a target sound from a mixed audio mixture using text labels.

## Datasets

- **ESC-50** — total 2000; splits: test (2000)

## Metrics

- `SDRi` **(primary)** — range: dB
  - Signal-to-distortion ratio improvement, calculated as the difference between the SDR of the separated output and the original mixture.
- `SI-SDR` — range: dB
  - Scale-invariant signal-to-distortion ratio, evaluates separation quality independent of amplitude scaling between prediction and target.

## Input / output format

**Input**: A 5-second audio mixture (SNR 0 dB) containing two clips from different semantic classes, paired with a text label query specifying the target class.

**Output**: Separated audio waveform corresponding to the target sound class specified by the text query.

## Scoring recipe

```python
def score(pred, gold, mixture):
  sdr_out = compute_sdr(pred, gold)
  sdr_mix = compute_sdr(mixture, gold)
  sdri = sdr_out - sdr_mix
  si_sdr = compute_si_sdr(pred, gold)
  return sdri, si_sdr
```

## Common pitfalls

- All clips are downsampled to 32 kHz for consistent evaluation, differing from the original 44.1 kHz.
- Mixtures are formed by randomly pairing two clips from different sound classes, not from the same class.

## Evidence (verbatim from paper)

> The ESC-50 dataset contains 2000 environmental audio recordings evenly arranged into 50 semantic classes. ... This leads to a total of 2000 evaluation pairs, which are used to evaluate the zero-shot performance of our model on environmental sound separation. ... We utilize signal-to-distortion ratio improvement (SDRi) and scale-invariant SDR (SI-SDR) to evaluate the performance of sound separation systems.

## Citation

```bibtex
@misc{liu2023separate,
  title={Separate Anything You Describe},
  author={Liu et al. (2023)},
  year={2023},
  note={arXiv:2308.05037}
}
```

- arXiv: 2308.05037