Evaluate DAMOS under natural localization errors

Evaluate DAMOS under the localization model’s naturally occurring error distribution on real, non-synthetic distortions to directly assess its practical robustness.

Background

DAMOS currently uses synthetically generated distortion masks both to train its distortion-localization model and to perform inference-time robustness analysis. The authors identify a need to test the framework against the kinds of localization errors that arise naturally when processing real, non-synthetic speech distortions, because such evaluation would provide a more direct measure of deployment robustness.

References

Despite these promising results, several directions remain open. First, the current framework relies on synthetically generated distortion masks for both training the localization model and conducting our inference-time robustness analysis; evaluating DAMOS under the localization model’s naturally occurring error distribution on real, non-synthetic distortions would provide a more direct measure of its practical robustness.

— DAMOS: Learning Distortion-Aware Speech Quality Assessment through Explicit Distortion Localization  (2608.21176 - Li et al., 21 Aug 2026) in Conclusion, p. 11

The detector is trained and evaluated on synthetically corrupted data; generalisation to naturally occurring translation errors remains unverified.

— Fact over Fiction: Detection of Pathological Hallucinations in Sinhala-to-English Neural Machine Translation  (2610.11389 - Obeysekara et al., 8 Oct 2026) in Limitations