MURMUR: An Efficient Inference System for Long-Form ASR

📰 ArXiv cs.AI

Learn how MURMUR improves long-form ASR with efficient inference, balancing accuracy and latency

advanced Published 2 Jun 2026

Action Steps

Implement MURMUR's chunk-based pipeline to process audio in parallel windows for low latency
Apply long-context ASR models to resolve cross-chunk context and improve accuracy
Configure MURMUR's system to balance accuracy and latency for specific use cases
Test MURMUR's performance on various long-form ASR tasks to evaluate its efficiency
Compare MURMUR's results with existing ASR systems to assess its improvements

Who Needs to Know This

Speech recognition engineers and researchers can benefit from MURMUR's approach to improve ASR systems, particularly those working on long-form audio processing

Key Insight

💡 MURMUR's approach combines the benefits of chunk-based pipelines and long-context ASR models to achieve efficient inference for long-form ASR

Key Takeaways

Learn how MURMUR improves long-form ASR with efficient inference, balancing accuracy and latency

Full Article

Title: MURMUR: An Efficient Inference System for Long-Form ASR

Abstract:
arXiv:2606.01483v1 Announce Type: cross Abstract: Long-form automatic speech recognition (ASR) requires both high accuracy and low latency, but existing systems force a trade-off between the two. Chunk-based pipelines process audio in parallel windows for low latency, but lose cross-chunk context and need brittle heuristics to align speakers and timestamps at boundaries. Long-context ASR models resolve everything in a single pass for better accuracy, but are an order of magnitude slower. We prop

Read full paper → ← Back to Reads