vi high-quality speech corpus

This dataset contains the guarded high-quality Vietnamese speech corpus and its source-partitioned audio.

All audio is FLAC PCM16. Source sample rate and channel count are preserved. Every packaged item passed a source-to-FLAC bit-identical PCM round trip and a separate full-release decode/hash validation.

Corpus rows use portable locators of the form audio/<subset>/<group>/part-N.parquet#row=N.

Corpus overview

Corpus/config	Rows	Approx. hours
vi	11,021,040	12,489.20
vi/validation	1,476	2.00

Audio subset overview

Counts below are unique packaged audio rows.

Audio subset	Unique rows	Approx. hours	Rejected rows	Parquet GiB
google_fleurs__vi_vn	4,181	13.09	0	0.60
leduckhai_multimed__vietnamese	5,247	9.13	0	0.30
linhtran92__viet_bud500	640,340	457.84	0	28.11
tensorxt_vimedcss__default	8,777	17.06	0	2.72
vietnamese_punctuated_jsonl__disco_eth_worldspeech	337,210	700.73	0	39.22
vietnamese_punctuated_jsonl__lemas_project_lemas_dataset_train	4,506,980	4,675.56	0	298.96
vietnamese_punctuated_jsonl__nhutp_vietspeech	617,054	694.58	0	43.30
vietnamese_speech_metadata__speechcolab_gigaspeech2	4,902,727	5,923.51	0	388.04

JSONL examples

Two abridged corpus rows are shown below; the locators point directly to the packaged Parquet audio rows.

{"audio_filepath":"audio/vietnamese_punctuated_jsonl__disco_eth_worldspeech/1ea93ffd627301f978b0/part-00000.parquet#row=0","text":"Còn bà Ngô Thị Bích Huyền thuộc tổ chức từ thiện Trẻ em Sài Gòn khi trả lời hãng tin Al Jazeera cho biết:","context":"","duration":6.71999979019165,"lang":"<|VI|>","audio_id":"ea2fb56858bce427c90097a143d580b15f2d69df","source_set":"vietnamese_punctuated_jsonl__disco_eth_worldspeech","audio_format":"flac"}
{"audio_filepath":"audio/vietnamese_punctuated_jsonl__disco_eth_worldspeech/1ea93ffd627301f978b0/part-00000.parquet#row=500","text":"sử dụng vốn ODA, nêu rõ ưu điểm cũng như các hạn chế yếu kém của các cơ quan hữu quan.","context":"","duration":7.0,"lang":"<|VI|>","audio_id":"22ce2d6380b25dd43cb2aebcca21811e1b95f027","source_set":"vietnamese_punctuated_jsonl__disco_eth_worldspeech","audio_format":"flac"}

Download only selected subsets

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="ACCOUNT/vi",
    repo_type="dataset",
    allow_patterns=[
        "corpus/*",
        "audio/google_fleurs__vi_vn/*/part-*.parquet",
    ],
)

Or load an individual configuration:

from datasets import load_dataset
corpus = load_dataset("ACCOUNT/vi", "vi")
audio = load_dataset("ACCOUNT/vi", "audio-google_fleurs__vi_vn")

Release totals

• Unique audio: 11,022,516
• Audio hours: 12,491.494
• Parquet size: 0.860330 TB (0.782465 TiB)

See release.json for exact per-subset counts, hours, bytes, and corpus checksums.