English

Efficient and Robust Long-Form Speech Recognition with Hybrid H3-Conformer

Sound 2024-10-08 v1 Audio and Speech Processing

Abstract

Recently, Conformer has achieved state-of-the-art performance in many speech recognition tasks. However, the Transformer-based models show significant deterioration for long-form speech, such as lectures, because the self-attention mechanism becomes unreliable with the computation of the square order of the input length. To solve the problem, we incorporate a kind of state-space model, Hungry Hungry Hippos (H3), to replace or complement the multi-head self-attention (MHSA). H3 allows for efficient modeling of long-form sequences with a linear-order computation. In experiments using two datasets of CSJ and LibriSpeech, our proposed H3-Conformer model performs efficient and robust recognition of long-form speech. Moreover, we propose a hybrid of H3 and MHSA and show that using H3 in higher layers and MHSA in lower layers provides significant improvement in online recognition. We also investigate a parallel use of H3 and MHSA in all layers, resulting in the best performance.

Keywords

Cite

@article{arxiv.2410.04159,
  title  = {Efficient and Robust Long-Form Speech Recognition with Hybrid H3-Conformer},
  author = {Tomoki Honda and Shinsuke Sakai and Tatsuya Kawahara},
  journal= {arXiv preprint arXiv:2410.04159},
  year   = {2024}
}

Comments

Submitted to InterSpeech2024, Sample code is available at https://github.com/mirrormouse/Hybrid-H3-Conformer

R2 v1 2026-06-28T19:09:45.313Z