metadata
configs:
- config_name: en
data_files:
- split: train
path: corpus/en.train.clean.jsonl
- split: eval
path: corpus/en.eval.clean.jsonl
- config_name: en-5000h
data_files:
- split: train
path: corpus/en.train.coverage-5000h.jsonl
- split: eval
path: corpus/en.eval.clean.jsonl
- config_name: audio-amphion_emilia_dataset__en
data_files:
- split: train
path: audio/amphion_emilia_dataset__en/*/part-*.parquet
- config_name: audio-disco_eth_worldspeech__en
data_files:
- split: train
path: audio/disco_eth_worldspeech__en/*/part-*.parquet
- config_name: audio-mozilla_common_voice_scripted_english__extracted_audio
data_files:
- split: train
path: >-
audio/mozilla_common_voice_scripted_english__extracted_audio/*/part-*.parquet
- config_name: audio-nvidia_granary__librilight
data_files:
- split: train
path: audio/nvidia_granary__librilight/*/part-*.parquet
- config_name: audio-nvidia_granary__librilight__extension_50k_v1
data_files:
- split: train
path: audio/nvidia_granary__librilight__extension_50k_v1/*/part-*.parquet
- config_name: audio-nvidia_granary__voxpopuli
data_files:
- split: train
path: audio/nvidia_granary__voxpopuli/*/part-*.parquet
- config_name: audio-nvidia_granary__voxpopuli__extension_50k_v1
data_files:
- split: train
path: audio/nvidia_granary__voxpopuli__extension_50k_v1/*/part-*.parquet
- config_name: audio-nvidia_granary__yodas
data_files:
- split: train
path: audio/nvidia_granary__yodas/*/part-*.parquet
- config_name: audio-nvidia_granary__yodas__extension_50k_v1
data_files:
- split: train
path: audio/nvidia_granary__yodas__extension_50k_v1/*/part-*.parquet
- config_name: audio-nvidia_granary__ytc
data_files:
- split: train
path: audio/nvidia_granary__ytc/*/part-*.parquet
- config_name: audio-voices_in_the_wild_jsonl__audio
data_files:
- split: train
path: audio/voices_in_the_wild_jsonl__audio/*/part-*.parquet
en cleaned speech corpus
This refresh provides a converged full train manifest, a frozen eval manifest, and a token-coverage subset of approximately 5000 hours. The validated FLAC payloads are reused without decoding or transcoding. Completely referenced Parquet shards are hard-linked locally; partial shards are filtered into exact row-only Parquet files.
Corpus overview
| Variant | Rows | Hours |
|---|---|---|
| train | 10,486,951 | 49951.119 |
| coverage_5000h | 972,368 | 4999.858 |
| eval | 544 | 1.999 |
All audio locators are portable and have the form
audio/<subset>/<group>/part-N.parquet#row=N.
The release audit records exact input/output hashes, rejected audio mappings, language-tag validation, and zero train/eval audio-identity overlap.
from datasets import load_dataset
ds = load_dataset("YuKuanFu/en", "en-5000h", split="train")
Packaging exclusions
Rows that could not be represented by the validated FLAC pool are omitted and recorded here rather than silently replaced:
| Variant | Rows | Hours | Reasons |
|---|---|---|---|
| train | 811 | 4.232495 | duration_mismatch: 811 |
| coverage_5000h | 27 | 0.142076 | duration_mismatch: 27 |