thelarryproject/libriheavy-test_clean_large-index-vocoder

thelarryproject Libriheavy test_clean_large Vocoder Index

An index over thelarryproject/libriheavy-test-clean-large-normalized-asr (Libriheavy's test_clean_large set): the clips that pass the audio quality and transcript filters used for vocoder training, as row numbers and ids only. Each row points at its clip in the source dataset by row number.

21,661 of the source's 23,464 clips remain:

split	clips	hours	speakers
train	21,661	81.0	70

Fields

• row: the clip's row number in the source's train split
• id: the source's clip id, e.g. large/9676/unhappy_far-off_things_1701_librivox_64kb_mp3/unhappyfaroffthings_07_dunsany_64kb_19; check it against the source row to confirm the match

Construction

Rows are numbered first. Removed, with the stricter thresholds used for synthesis data:

• clips shorter than 0.1 s or longer than 25 s
• clips with ctc_mismatch above 0.2
• clips failing an audio quality check: overall level below -40 dBFS, speech in under 50% of frames, signal-to-noise ratio below 15 dB, or more than 0.01% of samples clipped

Then every column but row and id is dropped. See provenance.json for the producing commit and timestamp.

Known corrections

N/a

Known gaps

• Only row and id: read audio, transcripts, and every other column from the source dataset by row; a later revision of the source may reorder rows
• Not speaker-disjoint from Libriheavy's training data: 31 of the source's 72 speakers also appear in the large config

Libriheavy is licensed Apache-2.0, and this redistribution is Apache-2.0 under thelarryproject. The underlying audio is public domain LibriVox recordings.

Citation

If you use this dataset, please cite the original corpus as well:

@misc{kang2023libriheavy,
  title         = {Libriheavy: a 50,000 hours ASR corpus with punctuation casing and context},
  author        = {Wei Kang and Xiaoyu Yang and Zengwei Yao and Fangjun Kuang and Yifan Yang and Liyong Guo and Long Lin and
                   Daniel Povey},
  year          = {2023},
  eprint        = {2309.08105},
  archivePrefix = {arXiv},
  primaryClass  = {eess.AS}
}

Preprocessing pipeline code is at https://github.com/Mekadrom/larry, and this dataset is fully reproducible with a single command in that repo:

uv run preprocessing -f configs/preprocessing/index/voice/vocoder/mythicinfinity_libriheavy-test_clean_large-index-vocoder.yaml --output_dir ~/.cache/larry/datasets/index/voice/vocoder/mythicinfinity_libriheavy-test_clean_large-index-vocoder --clean