# On the Role of Encoder Depth: Pruning Whisper and LoRA Fine-Tuning in SLAM-ASR

arXiv: https://arxiv.org/abs/2603.27981 · HTML: https://mlsanity.com/paper/2603.27981/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 18 of its 26 references that are in the map.
- First posted on arXiv 30 Mar 2026.

## Builds on (most cited first, 18 of 18)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Parameter-Efficient Transfer Learning for NLP](https://mlsanity.com/paper/1902.00751/) (2019)
- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [Common Voice: A Massively-Multilingual Speech Corpus](https://mlsanity.com/paper/1912.06670/) (2019)
- [Layer-wise Analysis of a Self-supervised Speech Representation Model](https://mlsanity.com/paper/2107.04734/) (2021)
- [On decoder-only architecture for speech-to-text and large language model integration](https://mlsanity.com/paper/2307.03917/) (2023)
- [Distil-Whisper: Robust Knowledge Distillation via Large-Scale Pseudo Labelling](https://mlsanity.com/paper/2311.00430/) (2023)
- [An Embarrassingly Simple Approach for LLM with Strong ASR Capacity](https://mlsanity.com/paper/2402.08846/) (2024)
- [LoRA-Whisper: Parameter-Efficient and Extensible Multilingual ASR](https://mlsanity.com/paper/2406.06619/) (2024)
- [HDMoLE: Mixture of LoRA Experts with Hierarchical Routing and Dynamic Thresholds for Fine-Tuning LLM-based ASR Models](https://mlsanity.com/paper/2409.19878/) (2024)
- [Text-only adaptation in LLM-based ASR through text denoising](https://mlsanity.com/paper/2601.20900/) (2026)
- [Low-Resource Domain Adaptation for Speech LLMs via Text-Only Fine-Tuning](https://mlsanity.com/paper/2506.05671/) (2025)
- [Speech LLMs in Low-Resource Scenarios: Data Volume Requirements and the Impact of Pretraining on High-Resource Languages](https://mlsanity.com/paper/2508.05149/) (2025)
- [The Eloquence team submission for task 1 of MLC-SLM challenge](https://mlsanity.com/paper/2507.19308/) (2025)
- [BaldWhisper: Faster Whisper with Head Shearing and Layer Merging](https://mlsanity.com/paper/2510.08599/) (2025)
- [Pruning as Regularization: Sensitivity-Aware One-Shot Pruning in ASR](https://mlsanity.com/paper/2511.08092/) (2025)

## Built on by (most cited first, 2 of 2)

- [X-AuT: Progressive Audio-Encoder Compression for Speech LLMs with Cross-Scale Distillation](https://mlsanity.com/paper/2609.11412/) (2026)
- [Pruning for Efficiency, Paying in Fairness: Demographic Disparities in Pruned Speech-LLMs](https://mlsanity.com/paper/2609.38106/) (2026)

## Abstract (arXiv)

Automatic speech recognition (ASR) has advanced rapidly in recent years, driven by large-scale pretrained models and end-to-end architectures such as SLAM-ASR. A key component of SLAM-ASR systems is the Whisper speech encoder, which provides robust acoustic representations. While model pruning has been explored for the full Whisper encoder-decoder architecture, its impact within the SLAM-ASR setting remains under-investigated. In this work, we analyze the effects of layer pruning in the Whisper encoder when used as the acoustic backbone of SLAM-ASR. We further examine the extent to which LoRA-based fine-tuning can recover performance degradation caused by pruning. Experiments conducted across three Whisper variants (Small, Medium, Large-v2), three languages representing distinct resource levels (Danish, Dutch, English), and over 200 training runs demonstrate that pruning two encoder layers causes only 2-4% WER degradation, and that combining this pruning with LoRA adaptation consistently outperforms the unpruned baseline while reducing total parameters by 7-14%. Moreover, our error analysis reveals that LoRA primarily compensates through the language model's linguistic priors, reducing total word errors by 11-21% for Dutch and English, with substitutions and deletions showing the largest reductions. However, for low-resource Danish, the reduction is smaller (4-7%), and LoRA introduces increased insertion errors, indicating that compensation effectiveness depends on the LLM's pre-existing language proficiency and available training data.