A novel multi-head efficient decoding approach for transformer-based Automatic Speech Recognition (ASR), improving inference speed and accuracy.