Included docs
What ships in the public metadata surface
cantharm_release_readme.mdcantharm_release_manifest.jsoncantharm_dataset_statistics.csvcantharm_benchmark_highlights.csvcantharm_reproducibility_note.mdcantharm_audio_inventory.csv
Manifest summary
Core bundle facts
| Field | Value |
|---|---|
| Release name | cantharm_public_release_bundle |
| Release snapshot | 2026-04-02 |
| Created on | 2026-04-02 |
| Audio files | 4,823 canonical .wav files |
| Audio naming pattern | Fxxxxx_SPKyy.wav |
| Speaker ids | SPK01 to SPK11 |
| Total audio size | 1,903,438,994 bytes |
Explicit exclusions
What does not ship in the public bundle
- Derived CSV exports and split files
- Trained checkpoints and intermediate result directories
- Non-release staging materials
- Internal adjudication notes, staging labels, and recording logs
Website-hosted public package
Download surface used by this release
This site serves stable public filenames rather than internal staging names. The public workbook, manifest, concise statistics, benchmark summary, reproducibility note, and speaker-level audio packages are exposed from Downloads.
The benchmark summary shown on this site uses the same release snapshot and public filenames as the manuscript.
Workbook rule
How the public workbook should be cited
The site, paper, and download bundle all treat cantharm_release_workbook.xlsx as the public workbook name for the locked release snapshot. Public-facing references on this site use that alias consistently.
The release manifest, metadata bundle, and checksum on this page are the source of truth for the public handoff. Earlier staging names and intermediate mirrors are intentionally not part of the public package surface.