Derived from the existing Namecoach/netjets-name-asr train split and the HEDB_Conf worksheet in NetJets Submission Post-Processing_ 2_11_22 - 04_23.xlsx.

This review-safe dataset keeps all 36,704 existing HF train rows and only the requested columns: Name, embedded playable audio, Final Phonetic, phonetic_system, n_total, and perc. Audio is stored as bytes with audio.path = null, so it does not depend on a local source path.

n_total/perc use an exact full-name + phonetic match from HEDB_Conf when available; otherwise they are deterministically derived from the complete HF train split by surname and pronunciation variant. This keeps every existing HF row usable while preserving the workbook context where it is unambiguous.

The Argilla review queue uses exactly 1,669 deterministic rows selected across pronunciation ambiguity, coverage bands, and surname groups. Most selected surname groups are kept complete so alternate recorded forms stay together.

Downloads last month: 27

Total file size:

2.72 GB