FastSLM: Hierarchical Temporal Abstraction for Efficient Long-Form Speech Adaptation
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911737574326272 |
|---|---|
| author | Lee, Junseok Lee, Sangyong Chun, Chang-Jae |
| author_facet | Lee, Junseok Lee, Sangyong Chun, Chang-Jae |
| contents | Scaling Multimodal Large Language Models (MLLMs) to long-form speech is bottlenecked by the explosive growth of input tokens. Unlike images or videos, audio lacks overlapping information, making extreme 1-token compression highly susceptible to the loss of fine-grained acoustic cues. To overcome this, we propose FastSLM, a token-efficient architecture featuring the Hierarchical Temporal Abstractor (HTA). HTA progressively distills non-overlapping acoustic features across multiple temporal scales, achieving an extreme compression rate of 1.67 tokens per second a 97% reduction without losing critical context. Experimental results show that FastSLM achieves competitive performance with state-of-the-art models on long-form benchmarks despite operating with significantly fewer FLOPs and parameters. The source code and model checkpoints are available at https://anonymous.4open.science/r/FastSLM-8BD3. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_06199 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | FastSLM: Hierarchical Temporal Abstraction for Efficient Long-Form Speech Adaptation Lee, Junseok Lee, Sangyong Chun, Chang-Jae Audio and Speech Processing Artificial Intelligence Sound Scaling Multimodal Large Language Models (MLLMs) to long-form speech is bottlenecked by the explosive growth of input tokens. Unlike images or videos, audio lacks overlapping information, making extreme 1-token compression highly susceptible to the loss of fine-grained acoustic cues. To overcome this, we propose FastSLM, a token-efficient architecture featuring the Hierarchical Temporal Abstractor (HTA). HTA progressively distills non-overlapping acoustic features across multiple temporal scales, achieving an extreme compression rate of 1.67 tokens per second a 97% reduction without losing critical context. Experimental results show that FastSLM achieves competitive performance with state-of-the-art models on long-form benchmarks despite operating with significantly fewer FLOPs and parameters. The source code and model checkpoints are available at https://anonymous.4open.science/r/FastSLM-8BD3. |
| title | FastSLM: Hierarchical Temporal Abstraction for Efficient Long-Form Speech Adaptation |
| topic | Audio and Speech Processing Artificial Intelligence Sound |
| url | https://arxiv.org/abs/2601.06199 |