# Six Layers Less: Encoder Pruning for Whisper with Label-Free Recovery

arXiv: https://arxiv.org/abs/2609.27980 · HTML: https://mlsanity.com/paper/2609.27980/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 10 of its 12 references that are in the map.
- First posted on arXiv 23 Sep 2026.

## Builds on (most cited first, 10 of 10)

- [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)
- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech](https://mlsanity.com/paper/2205.12446/) (2022)
- [The Unreasonable Ineffectiveness of the Deeper Layers](https://mlsanity.com/paper/2403.17887/) (2024)
- [The People's Speech: A Large-Scale Diverse English Speech Recognition Dataset for Commercial Usage](https://mlsanity.com/paper/2111.09344/) (2021)
- [Distil-Whisper: Robust Knowledge Distillation via Large-Scale Pseudo Labelling](https://mlsanity.com/paper/2311.00430/) (2023)
- [LiteASR: Efficient Automatic Speech Recognition with Low-Rank Approximation](https://mlsanity.com/paper/2502.20583/) (2025)
- [WhisperKit: On-device Real-time ASR with Billion-Scale Transformers](https://mlsanity.com/paper/2507.10860/) (2025)
- [Rethinking Layer Relevance in Large Language Models Beyond Cosine Similarity](https://mlsanity.com/paper/2605.14075/) (2026)

## Abstract (arXiv)

Pruning large pre-trained transformer-based ASR models such as OpenAI's Whisper has seen great adoption, as pruning the decoder led to significant end-to-end transcription speedups. For instance, the tt whisper-large-v3-turbo variant reduced the decoder from 32 to 4 layers, while Distill-Whisper similarly reduced the decoder to only 2 layers. Although some attention has been put towards reducing the size of the encoder, no approach has seen wide adoption. This could be due to the need for custom inference implementations to take advantage of the compressed model. We present an approach that ranks encoder layers by the leave-one-layer-out change in Word Error Rate (WER). The six layers that cause the least change are removed, corresponding to 18.5% of the encoder stack. The pruned model requires no custom inference code as it is simply a more shallow encoder with fewer layers. We further distill using unlabeled monolingual speech data to recover performance degradation caused by the zero-shot layer pruning. Mean WER across four languages increases to 20.1% after distillation, compared to 21.9% zero-shot, going from a baseline of 18.2%. We release all of our code (https://github.com/rasgaard/whisper-encoder-layer-prune) and the pruned model (https://huggingface.co/rasgaard/whisper-large-v3-turbo-encoder-pruned).