Datasetpublic

BrunoHays/apollo-telephony-en

Apollo telephony — English (processed) English Apollo air-to-ground / mission-control style telephony speech from the Apollo FSC P4 ASR track-2 release (WAV + transcriptions.txt in apollo_data.zip). Credits & license Original corpus: Apollo FSC P4 ASR track 2.License: CC BY-4.0 (commercial use allowed with attribution). Snapshot Language: en Samples: 48068 clips (~38.39 h) Audio: OGG/Opus mono, native 8 kHz telephony Construction: utterance-level… See the full description on the dataset page: https://huggingface.co/datasets/BrunoHays/apollo-telephony-en.

sourceHugging Facecc-by-4.0updated 1d agoView on Hugging Face

0likes25downloads

Dataset Card

Apollo telephony — English (processed)

English Apollo air-to-ground / mission-control style telephony speech from the Apollo FSC P4 ASR track-2 release (WAV + transcriptions.txt in apollo_data.zip).

Credits & license

Original corpus: Apollo FSC P4 ASR track 2. License: CC BY-4.0 (commercial use allowed with attribution).

Snapshot

—Language: en
—Samples: 48068 clips (~38.39 h)
—Audio: OGG/Opus mono, native 8 kHz telephony
—Construction: utterance-level source (no call-id merge). Upstream train/dev transcripts pooled into an eval-only remaster — every row split: "test". Ordered longest → shortest, full corpus (no hour cap). Upstream eval WAVs without transcripts are excluded. Upstream train/dev tokens may appear inside audio_id only (identity), not as Hub splits.
—License: CC-BY-4.0

Processing applied

1.Remaster to OGG/Opus mono at native 8 kHz telephony.
2.Schema columns: original_transcription, processed_transcription, transcription.
3.GT remediation: map [unk] → <unintelligible> on 4810 rows (processed_transcription / transcription). All other wording, ALL CAPS, and spelled-out numerals (ONE ZERO, NINER, pad readbacks) are kept as-is — that orthography is intentional Apollo style.
4.original_transcription remains the upstream text (immutable).

Dataset structure

field	description
`audio`	OGG/Opus, 8 kHz mono telephony
`audio_id`	e.g. `apollo-telephony-en_fsc_p4_ASR_track2_train_34911`
`license`	`CC-BY-4.0`
`original_transcription`	upstream transcript (immutable)
`processed_transcription`	tag-normalized text, or empty if unchanged
`transcription`	best available (= processed if set, else original)
`dur` / `sample_rate` / `lang` / `split` / `src` / `channels`	metadata

Notes

Mission-control / loop terminology with spelled letters and numerals. High WER against modern ASR is often orthography / ITN, not wrong labels.