Exploring the potential and limitations of Model Merging for Multi-Domain Adaptation in ASR

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Carvalho, Carlos, Teixeira, Francisco, Rolland, Thomas, Abad, Alberto
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917316226187264
author Carvalho, Carlos
Teixeira, Francisco
Rolland, Thomas
Abad, Alberto
author_facet Carvalho, Carlos
Teixeira, Francisco
Rolland, Thomas
Abad, Alberto
contents Model merging is a scalable alternative to multi-task training that combines the capabilities of multiple specialised models into a single model. This is particularly attractive for large speech foundation models, which are typically adapted through domain-specific fine-tuning, resulting in multiple customised checkpoints, for which repeating full fine-tuning when new data becomes available is computationally prohibitive. In this work, we study model merging for multi-domain ASR and benchmark 11 merging algorithms for 10 European Portuguese domains, evaluating in-domain accuracy, robustness under distribution shift, as well as English and multilingual performance. We further propose BoostedTSV-M, a new merging algorithm based on TSV-M that mitigates rank collapse via singular-value boosting and improves numerical stability. Overall, our approach outperforms full fine-tuning on European Portuguese while preserving out-of-distribution generalisation in a single model.
format Preprint
id arxiv_https___arxiv_org_abs_2603_05354
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Exploring the potential and limitations of Model Merging for Multi-Domain Adaptation in ASR
Carvalho, Carlos
Teixeira, Francisco
Rolland, Thomas
Abad, Alberto
Computation and Language
Audio and Speech Processing
Model merging is a scalable alternative to multi-task training that combines the capabilities of multiple specialised models into a single model. This is particularly attractive for large speech foundation models, which are typically adapted through domain-specific fine-tuning, resulting in multiple customised checkpoints, for which repeating full fine-tuning when new data becomes available is computationally prohibitive. In this work, we study model merging for multi-domain ASR and benchmark 11 merging algorithms for 10 European Portuguese domains, evaluating in-domain accuracy, robustness under distribution shift, as well as English and multilingual performance. We further propose BoostedTSV-M, a new merging algorithm based on TSV-M that mitigates rank collapse via singular-value boosting and improves numerical stability. Overall, our approach outperforms full fine-tuning on European Portuguese while preserving out-of-distribution generalisation in a single model.
title Exploring the potential and limitations of Model Merging for Multi-Domain Adaptation in ASR
topic Computation and Language
Audio and Speech Processing
url https://arxiv.org/abs/2603.05354