ru 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	4,650,979	4465.559
coverage_5000h	4,304,751	3879.927
eval	1,563	2.013

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/ru", "ru-5000h", split="train")

Coverage-size note

The completed strict eligible pool contains 3,880.000 source hours and could not reach 5,000 hours. After packaging exclusions, the uploaded ru-5000h train contains 3,879.927 hours. The default configuration uses the larger conservative full-clean train manifest.

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	70	0.073172	all_zero_audio: 70
coverage_5000h	70	0.073172	all_zero_audio: 70