@article{attentionorconvolutiontransformerencoder,
  title = {Attention or Convolution: Transformer Encoders in Audio Language Models for Inference Efficiency},
  author = {Sungho Jeon and Ching-Feng Yeh and Hakan Inan and Wei-Ning Hsu and Rashi Rungta and Yashar Mehdad and Daniel Bikel},
  year = {2023},
  eprint = {2311.02772},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2311.02772v2},
}