| --- |
| configs: |
| - config_name: vi |
| data_files: |
| - split: train |
| path: corpus/vi.train.clean.jsonl |
| - split: eval |
| path: corpus/vi.eval.clean.jsonl |
| - config_name: vi-5000h |
| data_files: |
| - split: train |
| path: corpus/vi.train.coverage-5000h.jsonl |
| - split: eval |
| path: corpus/vi.eval.clean.jsonl |
| - config_name: audio-google_fleurs__vi_vn |
| data_files: |
| - split: train |
| path: audio/google_fleurs__vi_vn/*/part-*.parquet |
| - config_name: audio-leduckhai_multimed__vietnamese |
| data_files: |
| - split: train |
| path: audio/leduckhai_multimed__vietnamese/*/part-*.parquet |
| - config_name: audio-linhtran92__viet_bud500 |
| data_files: |
| - split: train |
| path: audio/linhtran92__viet_bud500/*/part-*.parquet |
| - config_name: audio-tensorxt_vimedcss__default |
| data_files: |
| - split: train |
| path: audio/tensorxt_vimedcss__default/*/part-*.parquet |
| - config_name: audio-vietnamese_punctuated_jsonl__disco_eth_worldspeech |
| data_files: |
| - split: train |
| path: audio/vietnamese_punctuated_jsonl__disco_eth_worldspeech/*/part-*.parquet |
| - config_name: audio-vietnamese_punctuated_jsonl__lemas_project_lemas_dataset_train |
| data_files: |
| - split: train |
| path: audio/vietnamese_punctuated_jsonl__lemas_project_lemas_dataset_train/*/part-*.parquet |
| - config_name: audio-vietnamese_punctuated_jsonl__nhutp_vietspeech |
| data_files: |
| - split: train |
| path: audio/vietnamese_punctuated_jsonl__nhutp_vietspeech/*/part-*.parquet |
| - config_name: audio-vietnamese_speech_metadata__speechcolab_gigaspeech2 |
| data_files: |
| - split: train |
| path: audio/vietnamese_speech_metadata__speechcolab_gigaspeech2/*/part-*.parquet |
| --- |
| |
| # vi 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 | 11,017,247 | 12483.122 | |
| | coverage_5000h | 4,166,867 | 5000.000 | |
| | eval | 1,476 | 1.998 | |
| |
| 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. |
| |
| ```python |
| from datasets import load_dataset |
| ds = load_dataset("YuKuanFu/vi", "vi-5000h", split="train") |
| ``` |
| |
