Metraipolis › archive › dc_efeb0fe23ccc

CAL-MOS: Bridging Layers with Adapters for Robust MOS Prediction Across Speech Foundation Models — arXiv:2609.14956

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.14956
License (as archived): CC0-1.0
Archived: 2026-10-15T22:53:28.031Z
SHA-256 of the archived text: 80d957e8fa56bacbdf6732228523f047e325821407863e84b49a227bf864035e
Stable URL of this page: https://metraipolis.com/documento/dc_efeb0fe23ccc

How to cite

CAL-MOS: Bridging Layers with Adapters for Robust MOS Prediction Across Speech Foundation Models — arXiv:2609.14956 — arXiv (en), https://arxiv.org/abs/2609.14956 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_efeb0fe23ccc (Metraipolis archive dc_efeb0fe23ccc, sha256 80d957e8fa56bacbdf6732228523f047e325821407863e84b49a227bf864035e)

BibTeX

@misc{dcefeb0fe23ccc,
  title = {CAL-MOS: Bridging Layers with Adapters for Robust MOS Prediction Across Speech Foundation Models — arXiv:2609.14956},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.14956},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_efeb0fe23ccc; sha256 80d957e8fa56bacbdf6732228523f047e325821407863e84b49a227bf864035e},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.14956. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Speech Quality Assessment (SQA) is essential for modern speech technologies, and recent non-intrusive SQA predictors increasingly rely on Speech Foundation Models (SFMs). However, because SFMs expose representations from many layers, it remains unclear which depths are most informative for MOS prediction and how multi-layer information should be combined reliably across backbones and datasets. We benchmark ten SFMs on four MOS datasets under three regimes: full fine-tuning, last-layer probing with a frozen encoder, and naive cross-layer weighted aggregation. We find that the best layer is strongly backbone- and dataset-dependent, and that naive weighted fusion can be unstable across settings. We further evaluate a layer-calibrated aggregation variant that applies per-layer adapters before pooling, which improves the robustness of multi-layer fusion and narrows the gap to full fine-tuning while keeping the backbone frozen.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-15 in cs.AI).
Categorie arXiv: Sound (cs.SD) ; Artificial Intelligence (cs.AI); Audio and Speech Processing (eess.AS).
Pagina del lavoro: https://arxiv.org/abs/2609.14956 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-15T22:29:03.759Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-15?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).