Distilling Conversations: Abstract Compression of Conversational Audio Context for LLM-based ASR
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866915894904487936 |
|---|---|
| author | Kumar, Shashi Villatoro-Tello, Esaú Burdisso, Sergio Hacioglu, Kadri Bañeras-Roux, Thibault Watawana, Hasindri Sanchez-Cortes, Dairazalia Madikeri, Srikanth Motlicek, Petr Stolcke, Andreas |
| author_facet | Kumar, Shashi Villatoro-Tello, Esaú Burdisso, Sergio Hacioglu, Kadri Bañeras-Roux, Thibault Watawana, Hasindri Sanchez-Cortes, Dairazalia Madikeri, Srikanth Motlicek, Petr Stolcke, Andreas |
| contents | Standard LLM-based speech recognition systems typically process utterances in isolation, limiting their ability to leverage conversational context. In this work, we study whether multimodal context from prior turns improves LLM-based ASR and how to represent that context efficiently. We find that, after supervised multi-turn training, conversational context mainly helps with the recognition of contextual entities. However, conditioning on raw context is expensive because the prior-turn audio token sequence grows rapidly with conversation length. To address this, we propose Abstract Compression, which replaces the audio portion of prior turns with a fixed number of learned latent tokens while retaining corresponding transcripts explicitly. On both in-domain and out-of-domain test sets, the compressed model recovers part of the gains of raw-context conditioning with a smaller prior-turn audio footprint. We also provide targeted analyses of the compression setup and its trade-offs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_26246 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Distilling Conversations: Abstract Compression of Conversational Audio Context for LLM-based ASR Kumar, Shashi Villatoro-Tello, Esaú Burdisso, Sergio Hacioglu, Kadri Bañeras-Roux, Thibault Watawana, Hasindri Sanchez-Cortes, Dairazalia Madikeri, Srikanth Motlicek, Petr Stolcke, Andreas Computation and Language Artificial Intelligence Machine Learning Audio and Speech Processing Standard LLM-based speech recognition systems typically process utterances in isolation, limiting their ability to leverage conversational context. In this work, we study whether multimodal context from prior turns improves LLM-based ASR and how to represent that context efficiently. We find that, after supervised multi-turn training, conversational context mainly helps with the recognition of contextual entities. However, conditioning on raw context is expensive because the prior-turn audio token sequence grows rapidly with conversation length. To address this, we propose Abstract Compression, which replaces the audio portion of prior turns with a fixed number of learned latent tokens while retaining corresponding transcripts explicitly. On both in-domain and out-of-domain test sets, the compressed model recovers part of the gains of raw-context conditioning with a smaller prior-turn audio footprint. We also provide targeted analyses of the compression setup and its trade-offs. |
| title | Distilling Conversations: Abstract Compression of Conversational Audio Context for LLM-based ASR |
| topic | Computation and Language Artificial Intelligence Machine Learning Audio and Speech Processing |
| url | https://arxiv.org/abs/2603.26246 |