English

Multiresolution and Multimodal Speech Recognition with Transformers

Audio and Speech Processing 2020-05-01 v1 Computer Vision and Pattern Recognition Machine Learning Sound Machine Learning

Abstract

This paper presents an audio visual automatic speech recognition (AV-ASR) system using a Transformer-based architecture. We particularly focus on the scene context provided by the visual information, to ground the ASR. We extract representations for audio features in the encoder layers of the transformer and fuse video features using an additional crossmodal multihead attention layer. Additionally, we incorporate a multitask training criterion for multiresolution ASR, where we train the model to generate both character and subword level transcriptions. Experimental results on the How2 dataset, indicate that multiresolution training can speed up convergence by around 50% and relatively improves word error rate (WER) performance by upto 18% over subword prediction models. Further, incorporating visual information improves performance with relative gains upto 3.76% over audio only models. Our results are comparable to state-of-the-art Listen, Attend and Spell-based architectures.

Keywords

Cite

@article{arxiv.2004.14840,
  title  = {Multiresolution and Multimodal Speech Recognition with Transformers},
  author = {Georgios Paraskevopoulos and Srinivas Parthasarathy and Aparna Khare and Shiva Sundaram},
  journal= {arXiv preprint arXiv:2004.14840},
  year   = {2020}
}

Comments

Accepted for ACL 2020

R2 v1 2026-06-23T15:12:54.558Z