VoxZip: Semantic-Anchored Temporal KV Cache Compression for Long-Context Audio Inference
Open Access
- 9 November 2026
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 3667-3676
Abstract
No abstract availableFunding Information
- National Natural Science Foundation of China under Grant (No.U25B2064)
- the “Pioneer” and “Leading Goose” R&D Program of Zhejiang (No. 2025C02110)
- the Yongjiang Talent Project of Ningbo (No. 2024A-161-G)
This publication has 14 references indexed in Scilit:
- X-OPD: Cross-Modal On-Policy Distillation for Capability Alignment in Speech LLMs2026
- A Survey on Speech Large Language Models for UnderstandingIEEE Journal of Selected Topics in Signal Processing, 2025
- KV Cache Compression for Inference Efficiency in LLMs: A Review2025
- SpeechPrune: Context-Aware Token Pruning for Speech Information Retrieval2025
- DyCoke : Dynamic Compression of Tokens for Fast Video Large Language Models2025
- PACHAT: Persona-Aware Speech Assistant for Multi-party Dialogue2025
- Recent Advances in Speech Language Models: A Survey2025
- An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models2024
- LOOK-M: Look-Once Optimization in KV Cache for Efficient Multimodal Long-Context Inference2024
- H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models2023