Closing the Speech-Text Gap with Limited Audio for Effective Domain Adaptation in LLM-Based ASR
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917389647478784 |
|---|---|
| author | Bañeras-Roux, Thibault Burdisso, Sergio Villatoro-Tello, Esaú Sánchez-Cortés, Dairazalia Liu, Shiran Baroudi, Severin Kumar, Shashi Watawana, Hasindri E, Manjunath K Hacioglu, Kadri Motlicek, Petr Stolcke, Andreas |
| author_facet | Bañeras-Roux, Thibault Burdisso, Sergio Villatoro-Tello, Esaú Sánchez-Cortés, Dairazalia Liu, Shiran Baroudi, Severin Kumar, Shashi Watawana, Hasindri E, Manjunath K Hacioglu, Kadri Motlicek, Petr Stolcke, Andreas |
| contents | Conventional end-to-end automatic speech recognition (ASR) systems rely on paired speech-text data for domain adaptation. Recent LLM-based ASR architectures connect a speech encoder to a large language model via a projection module, enabling adaptation with text-only data. However, this introduces a modality gap, as the LLM is not exposed to the noisy representations produced by the speech projector. We investigate whether small amounts of speech can mitigate this mismatch. We compare three strategies: text-only adaptation, paired speech-text adaptation, and mixed batching (MB), which combines both. Experiments in in-domain and out-of-domain settings show that even limited speech consistently improves performance. Notably, MB using only 10% of the target-domain (less than 4 hours) speech achieves word error rates comparable to, or better than, conventional ASR fine-tuning with the full dataset, indicating that small amounts of speech provide a strong modality-alignment signal. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_06487 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Closing the Speech-Text Gap with Limited Audio for Effective Domain Adaptation in LLM-Based ASR Bañeras-Roux, Thibault Burdisso, Sergio Villatoro-Tello, Esaú Sánchez-Cortés, Dairazalia Liu, Shiran Baroudi, Severin Kumar, Shashi Watawana, Hasindri E, Manjunath K Hacioglu, Kadri Motlicek, Petr Stolcke, Andreas Computation and Language Conventional end-to-end automatic speech recognition (ASR) systems rely on paired speech-text data for domain adaptation. Recent LLM-based ASR architectures connect a speech encoder to a large language model via a projection module, enabling adaptation with text-only data. However, this introduces a modality gap, as the LLM is not exposed to the noisy representations produced by the speech projector. We investigate whether small amounts of speech can mitigate this mismatch. We compare three strategies: text-only adaptation, paired speech-text adaptation, and mixed batching (MB), which combines both. Experiments in in-domain and out-of-domain settings show that even limited speech consistently improves performance. Notably, MB using only 10% of the target-domain (less than 4 hours) speech achieves word error rates comparable to, or better than, conventional ASR fine-tuning with the full dataset, indicating that small amounts of speech provide a strong modality-alignment signal. |
| title | Closing the Speech-Text Gap with Limited Audio for Effective Domain Adaptation in LLM-Based ASR |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2604.06487 |