Researchers have introduced a novel method for optimizing OpenAI's Whisper Automatic Speech Recognition (ASR) models by specifically targeting the encoder architecture. While previous optimizations like whisper-large-v3-turbo focused on reducing the decoder, this approach addresses the encoder by identifying and removing the least impactful layers. The authors employed a ranking mechanism based on the leave-one-layer-out change in Word Error Rate (WER) to select the optimal layers for removal. By pruning the six layers that caused the smallest performance fluctuations, the researchers successfully reduced the encoder stack by 18.5%. To mitigate the performance loss inherent in zero-shot layer pruning, the team utilized distillation with unlabeled monolingual speech data. This recovery process improved the mean WER across four tested languages from 21.9% in the zero-shot pruned model to 20.1%, moving closer to the original baseline of 18.2%. Notably, this method produces a shallower encoder that functions with standard inference implementations, avoiding the need for specialized custom code. The resulting model offers a more efficient computational footprint while maintaining high transcription accuracy through label-free recovery techniques.