qhjqhj00 clotho-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 diverse environmental sounds using natural language captions. The benchmark tests isolation of a target sound from a concatenated background mixture.
Datasets
- Clotho v2 — total 5225; splits: test (5225)
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 15-30 second audio mixture (SNR 0 dB) formed by concatenating and truncating two clips, paired with one of five human-annotated captions.
Output: Separated audio waveform corresponding to the target sound described by the caption.
Scoring recipe
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
- Mixtures are created by concatenating two clips and truncating to the target length, not simple additive mixing.
- Each target clip is mixed with five different background clips, yielding 5 variations per original evaluation clip.
Evidence (verbatim from paper)
The Clotho v2 evaluation set includes 1045 audio clips, each provided with five human-annotated captions. ... This procedure culminates in a total of 5225 mixtures for evaluation. ... We utilize signal-to-distortion ratio improvement (SDRi) and scale-invariant SDR (SI-SDR) to evaluate the performance of sound separation systems.
Citation
@misc{liu2023separate,
title={Separate Anything You Describe},
author={Liu et al. (2023)},
year={2023},
note={arXiv:2308.05037}
}
- arXiv: 2308.05037