Datasets:
Dataset Viewer
The dataset viewer is not available for this subset.
Job has been terminated due to a temporary spike in resource usage and may be restarted later.
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Apollo telephony — English (processed)
English air-to-ground / mission-control telephony speech from the Apollo FSC P4 ASR track-2 corpus. Remastered for 8 kHz telephony ASR evaluation and training.
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, 8 kHz
- Splits: original Apollo partition —
train(36320 / ~29.76 h),dev(11748 / ~8.63 h) - License: CC-BY-4.0
What changed vs upstream
- Audio remastered to OGG/Opus mono at native 8 kHz.
- Light GT cleanup:
[unk]→<unintelligible>(4810 rows). Caps, spelled numerals (ONE ZERO,NINER), and other Apollo orthography are unchanged.
original_transcription is the upstream text; transcription is the working
reference (processed when set, else original).
Dataset structure
| field | description |
|---|---|
audio |
OGG/Opus, 8 kHz mono |
audio_id |
e.g. …_track2_train_34911 |
license |
CC-BY-4.0 |
original_transcription |
upstream transcript |
processed_transcription |
cleaned text when changed, else empty |
transcription |
best available (= processed if set, else original) |
dur / sample_rate / lang / split / src / channels |
metadata |
Notes
Spelled letters and numerals are intentional in this domain. High WER against modern ASR is often orthography / ITN, not wrong labels.
- Downloads last month
- 197