Build a pack

Fill in the metadata, then select your audio files — everything happens in your browser, nothing is uploaded. Download all three files, add them under packs/<your-id>/ in a clone of the runner repo, and open a pull request — see the field reference and sourcing-audio guide for what each field means and the exact submission steps.

GOESB never hosts audio — this builder packages metadata, a transcript manifest, and a content hash, not the audio bytes themselves. You still need somewhere for the audio to actually live (an existing open corpus, or your own hosting) before a pack is usable.

Identity

idRequired.versionprofile_idRequired.visibility

licenseRequired.

Sourcing audiosource url (optional)fetch_instructionsA runnable recipe, not prose — see the field guide for a worked example.

Metadata

languageRequired.dialect (optional)age_group (optional)recording_environmentRequired.microphone (optional)background_noise (optional)num_speakers (optional)speech_styletranscription_style (optional)

tags (optional, comma-separated)

Audio clips

Select the audio files themselves — nothing is uploaded, they're hashed and their duration read entirely in your browser. Each clip needs a reference transcript typed in below; there's no automatic transcription (an ASR-generated transcript can't be its own ground truth). The detected sample rate reflects your browser's own audio decoder, which may resample — double-check it against your source files if the exact rate matters.

Select at least one audio file.

Preview

Fill in every required field and every clip's transcript to enable the downloads.

The sha256 below is computed in your browser for immediate feedback — run python scripts/validate_assets.py locally before opening a PR. That's the same check CI runs, so if it passes there it won't fail there either.

pack.yaml

version: 1.0.0
visibility: open
metadata:
  speech_style: read
  tags: []

manifest.jsonl

// select audio files to generate this