Speechprune: Context-aware Token Pruning For Speech Information Retrieval

Abstract

We introduce Speech Information Retrieval (SIR), a new long-context task for Speech Large Language Models (Speech LLMs), and present SPIRAL, a 1,012-sample benchmark testing models' ability to extract critical details from approximately 90-second spoken inputs. While current Speech LLMs excel at short-form tasks, they struggle with the computational and representational demands of longer audio sequences. To address this limitation, we propose SpeechPrune, a training-free token pruning strategy that uses speech-text similarity and approximated attention scores to efficiently discard irrelevant tokens. In SPIRAL, SpeechPrune achieves accuracy improvements of 29% and up to 47% over the original model and the random pruning model at a pruning rate of 20%, respectively. SpeechPrune can maintain network performance even at a pruning level of 80%. This approach highlights the potential of token-level pruning for efficient and scalable long-form speech understanding.

Stats

• citations2
• S2 citations—
• github stars0
• HF likes0
• heat score3.58
• arxiv keylin2024speechprune

Related papers

• Structured Pruning Of Self-supervised Pre-trained Models For Speech Recognition And Understanding (2023)11.39
• Accurate And Structured Pruning For Efficient Automatic Speech Recognition (2023)7.81
• SPIRAL: Self-supervised Perturbation-invariant Representation Learning For Speech Pre-training (2022)5.56
• On The Interplay Between Sparsity, Naturalness, Intelligibility, And Prosody In Speech Synthesis (2021)5.24
• End-to-end Contrastive Language-speech Pretraining Model For Long-form Spoken Question Answering (2025)0.00
• SPADE: Structured Pruning And Adaptive Distillation For Efficient LLM-TTS (2025)0.00
• Distillation And Pruning For Scalable Self-supervised Representation-based Speech Quality Assessment (2025)8.09
• Fastadasp: Multitask-adapted Efficient Inference For Large Speech Language Model (2024)2.51