Amicus_LibriSpeech / README.md
OKHand's picture
Update README.md
79b823d verified
|
Raw
History Blame Contribute Delete
2.48 kB
metadata
dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: text
      dtype: string
configs:
  - config_name: default
    data_files:
      - split: train-clean-100
        path: train-clean-100/metadata.jsonl
      - split: train-clean-360
        path: train-clean-360/metadata.jsonl
      - split: train-other-500
        path: train-other-500/metadata.jsonl
      - split: dev-clean
        path: dev-clean/metadata.jsonl
      - split: dev-other
        path: dev-other/metadata.jsonl
      - split: test-clean
        path: test-clean/metadata.jsonl
      - split: test-other
        path: test-other/metadata.jsonl

Amicus_LibriSpeech

Self-contained Amicus speech dataset packaged for Hugging Face audiofolder.

Splits

  • train-clean-100: 28539 samples, 100.5905 hours
  • train-clean-360: 104014 samples, 363.6054 hours
  • train-other-500: 148686 samples, 496.8568 hours
  • dev-clean: 2694 samples, 5.3076 hours
  • dev-other: 2857 samples, 5.0578 hours
  • test-clean: 2611 samples, 5.3232 hours
  • test-other: 2932 samples, 5.2775 hours

Load with Hugging Face Datasets

from datasets import load_dataset

ds = load_dataset("audiofolder", data_dir=".")
print(ds)
print(ds["train-clean-100"][0]["audio"])

After uploading this folder to a dataset repository, replace data_dir="." with the dataset repo id if automatic builder detection works for your repository:

from datasets import load_dataset

ds = load_dataset("YOUR_NAMESPACE/Amicus_LibriSpeech")

Use with Amicus

Download the whole dataset repository, including Git LFS files, then launch Amicus training from the downloaded dataset root so relative audio_path values resolve correctly.

huggingface-cli download YOUR_NAMESPACE/Amicus_LibriSpeech \
  --repo-type dataset \
  --local-dir data/Amicus_LibriSpeech

cd data/Amicus_LibriSpeech
python /path/to/Amicus/training/stage1/1_semantic_alignment.py \
  --train_data train-clean-100.jsonl

Source

This dataset is based on LibriSpeech, available from OpenSLR: https://www.openslr.org/12

Citation

If you use this dataset, please cite the original LibriSpeech paper:

@inproceedings{panayotov2015librispeech,
  title={Librispeech: an ASR corpus based on public domain audio books},
  author={Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev},
  booktitle={Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on},
  pages={5206--5210},
  year={2015},
  organization={IEEE}
}