README.md

metadata

dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: duration
      dtype: float64
    - name: bam
      dtype: string
    - name: french
      dtype: string
    - name: asr-ctc
      dtype: string
    - name: asr-tdt
      dtype: string
    - name: asr-mt-ctc
      dtype: string
    - name: asr-mt-tdt
      dtype: string
    - name: st-ctc
      dtype: string
    - name: st-tdt
      dtype: string
    - name: lau-tdt-k1
      dtype: string
    - name: lau-ctc-k1
      dtype: string
    - name: lau-tdt-k5
      dtype: string
    - name: lau-ctc-k5
      dtype: string
    - name: lau-tdt-k0.2
      dtype: string
    - name: lau-ctc-k0.2
      dtype: string
    - name: lau-tdt-mse-k1
      dtype: string
    - name: lau-ctc-mse-k1
      dtype: string
    - name: cluster_label
      dtype: string
  splits:
    - name: test
      num_bytes: 121749438
      num_examples: 1218
  download_size: 117816418
  dataset_size: 121749438
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*

LAU eval dataset

This dataset was created while evaluating and comparing the models trained with Listen Attend Understand regularization and our E2E-ST model. The audio is from jeli-asr test set; the regularization loss weight lambda in the paper is represented by the character "k" in the fields of this dataset, each field represent a model with a specific decoding strategy (CTC or TDT)

---

Citation

@misc{diarra2026listenattendunderstandregularization,
      title={Listen, Attend, Understand: a Regularization Technique for Stable E2E Speech Translation Training on High Variance labels}, 
      author={Yacouba Diarra and Michael Leventhal},
      year={2026},
      eprint={2601.01121},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2601.01121}, 
}