bashkorttele/narrated-audiobooks-brsbs

Narrated Bashkir Audiobooks — BRSBS (Bashkorttele)

≈ 78.7 hours of human-narrated audiobooks, primarily in the Bashkir language, drawn from public-domain literary works and folk epics. Recorded as accessible "talking books" by the Bashkir Republican Special Library for the Blind (BRSBS) and released for language preservation and AI/ML research.

🌐 Languages of this card: English · Башҡортса · Русский

This dataset is part of a larger series published under the Bashkorttele project for the preservation and digital development of the Bashkir language.

---

English

Dataset summary

This dataset contains 476 MP3 files (≈ 78 h 40 m) covering 14 works — poetry, novels, novellas, folk tales, and traditional Bashkir epics (kubairs). All recordings are single-speaker or few-speaker studio narrations produced by the BRSBS as talking-book editions for blind and visually-impaired readers between 1989 and 2019.

The underlying literary works are in the public domain; the dataset (the compilation of recordings and metadata) is released under CDLA-Permissive-2.0.

At a glance

Works	14
Audio files	476
Total duration	78 h 40 m (measured from the audio)
Total size	≈ 4.17 GB
Primary language	Bashkir (ba)
Secondary	Russian (ru) — one bilingual work (Ural-batyr)
Audio format	MP3, 44.1 kHz, ~128–256 kbps (CBR/VBR by work)
Transcripts	Not included in this release (audio only)
License	CDLA-Permissive-2.0

Supported tasks

• Self-supervised / speech representation pretraining for a low-resource Turkic language.
• Automatic Speech Recognition (ASR) — as unlabeled pretraining audio, or after adding your own transcriptions (the source texts are public domain and can be aligned).
• Text-to-Speech (TTS) and voice research — clean, professional single-speaker narration per work.
• Language / speaker / genre audio classification.

⚠️ No transcriptions are shipped with this release. For supervised ASR you must supply aligned text. The corpus is otherwise directly usable for pretraining, TTS source material, and speech analysis.

Languages

Bashkir (ba), a Kipchak Turkic language of the Republic of Bashkortostan, Russian Federation. One work — the epic Ural-batyr — is narrated bilingually in Bashkir and Russian (ru). Language per file is given in the language column.

Dataset structure

narrated-audiobooks-brsbs/
├── README.md          # this dataset card
├── LICENSE            # full CDLA-Permissive-2.0 text
├── NOTICE             # attribution & provenance
├── metadata.csv       # one row per audio file (the metadata driver)
├── data/
│   ├── akbuzat/0001.mp3
│   ├── khairi-povorot/0001.mp3 … 0155.mp3
│   └── …              # one folder per work (14 total)
└── scripts/           # reproducible build & upload tooling (not part of the data)

Each work is a folder under data/ named with an ASCII slug; audio segments are zero-padded and ordered for playback (0001.mp3, 0002.mp3, …). The original Bashkir/Russian titles are preserved in the metadata, not in the file paths.

metadata.csv fields

Column	Description
file_name	Relative path to the MP3 (e.g. data/khairi-povorot/0001.mp3)
work_id	Slug identifier of the work
title_ru	Work title as printed in the source catalogue (Russian)
title_ba	Work title in Bashkir (Cyrillic)
title_en	Work title in English
author_native / author_ba / author_en	Author (folk works: "folk / халыҡ ижады")
author_life	Author's birth–death years
narrator / narrator_native	Reader (romanized / Cyrillic), where known
language	ba, or ba,ru for the bilingual work
work_type	epic / poems / poem / novel / novella / novellas / play / folk-tales
segment	1-based index of this file within its work
n_segments	Number of files in the work
duration_sec	Duration in seconds, measured from the audio
year	Year of the BRSBS recording/edition
publisher	BRSBS (recording institution)
source_edition	Original print edition, where noted
public_domain_source	true — underlying work is public domain
license	CDLA-Permissive-2.0
collection	Bashkorttele

How to load

from datasets import load_dataset

# From the Hub (audiofolder is auto-detected; metadata.csv columns are joined)
ds = load_dataset("bashkorttele/narrated-audiobooks-brsbs", split="train")

print(ds[0]["audio"])        # {'path': ..., 'array': ..., 'sampling_rate': 44100}
print(ds[0]["title_en"], ds[0]["work_type"], ds[0]["duration_sec"])

# Filter to a single work
povorot = ds.filter(lambda x: x["work_id"] == "khairi-povorot")

You can also load a local copy with load_dataset("audiofolder", data_dir="narrated-audiobooks-brsbs").

Works included

#	Author	Title (EN)	Bashkir title	Type	Narrator	Year	Files	Duration
1	folk	Akbuzat	Аҡбуҙат	epic	V. Kalmantaeva	2010	1	1:32:50
2	M. Akmulla (1831–1895)	Poems	Шиғырҙар	poems	—	2010	55	2:25:07
3	folk	Aldar and Zukhra	Алдар менән Зөһрә	epic	V. Kalmantaeva	2010	54	7:56:23
4	Sh. Babich (1895–1919)	Spring Song	Яҙғы йыр	poems	A. Sabiryanova	2010	71	3:16:34
5	folk	Magic Tales	Тылсымлы әкиәттәр	folk-tales	—	2014	21	4:49:27
6	M. Gafuri (1880–1934)	Poems	Шиғырҙар	poems	—	2011	2	4:11:27
7	M. Gafuri (1880–1934)	Red Star	Ҡыҙыл йондоҙ	play	—	2018	1	2:18:04
8	M. Gafuri (1880–1934)	Novellas	Повестар	novellas	—	2010	60	14:07:24
9	H. Davletshina (1905–1954)	Aybika	Айбикә	novella	A. Abizova	2019	7	2:34:26
10	H. Davletshina (1905–1954)	Irgiz	Ырғыҙ	novel	R. Karimova	1989	19	17:20:53
11	folk	Kara Yurga	Ҡара юрға	epic	—	2019	1	0:30:47
12	folk	Ural-batyr	Урал батыр	epic	—	2010	10	2:40:04
13	G. Khairi (1903–1938)	The Turn	Боролош	novel	A. Abizova	2010	155	14:23:03
14	S. Yulaev (1754–1800)	Poems	Шиғырҙар	poems	—	2010	19	0:33:46

Durations are measured from the audio and may differ slightly from the source catalogue.

Data collection and provenance

The recordings were produced by the Bashkir Republican Special Library for the Blind named after M. Kh. Tukhvatshin (BRSBS) in Ufa as accessible talking-book editions. They were compiled from the library's digital holdings and published for AI/ML use by the ANO for the Preservation and Development of the Bashkir Language under the Bashkorttele project. See NOTICE for full institutional attribution.

Files were losslessly reorganized for release: original recordings were renamed to ASCII slugs and re-numbered in playback order without re-encoding the audio. 19 byte-identical duplicate files (all in the Yulaev folder) were removed; no audio was transcoded or altered.

Each published MP3 also carries embedded ID3v2.4 tags (UTF-8) — title, work/album, author, narrator, track, year, publisher, language, copyright/license, plus custom TXXX fields (LICENSE, WORK_ID, TITLE_BA, TITLE_EN, AUTHOR_EN, NARRATOR, …) — so the files remain self-describing outside the Hub. The tags are prepended only; the audio stream is bit-for-bit identical to the source.

Licensing

• Dataset license: Community Data License Agreement – Permissive, Version 2.0 (CDLA-Permissive-2.0). The full text is in LICENSE. Per §2.1, this license text is distributed with the data.
• Underlying works: public domain (RF Civil Code Art. 1282(2)); folk epics/tales have no individual author.
• CDLA-Permissive-2.0 does not require attribution, but crediting the sources (see Citation) is appreciated.

Personal and sensitive information

The dataset contains professional literary narration only. It does not contain personal data about private individuals. Narrators are credited by professional name where known. No contracts, financial records, or identifying documents are included.

Considerations, limitations and biases

• No transcriptions in this release — supervised ASR requires user-supplied aligned text.
• Historical language: some works use early-20th-century or archaic Bashkir orthography and vocabulary.
• Speaker imbalance: a few narrators account for most of the audio; not balanced for speaker-robust modeling.
• Recording conditions vary across 1989–2019 (bitrate, room, equipment); this is real-world talking-book audio, not a controlled studio corpus.
• Segment boundaries follow the original talking-book structure (by track/chapter), not by sentence or utterance.

Citation

@misc{bashkorttele_brsbs_narrated_audiobooks_2026,
  title        = {Narrated Bashkir Audiobooks (BRSBS) — Bashkorttele},
  author       = {{ANO for the Preservation and Development of the Bashkir Language} and
                  {Bashkir Republican Special Library for the Blind named after M. Kh. Tukhvatshin}},
  year         = {2026},
  howpublished = {Hugging Face Hub},
  publisher    = {Bashkorttele},
  url          = {https://huggingface.co/datasets/bashkorttele/narrated-audiobooks-brsbs},
  note         = {Licensed under CDLA-Permissive-2.0}
}

Acknowledgements

Narrations by the BRSBS (Ufa). Published by the ANO for the Preservation and Development of the Bashkir Language.

---

Башҡортса

Дөйөм мәғлүмәт

Был датасетта башҡорт телендәге нәфис әҫәрҙәрҙең тауыш яҙмалары тупланған: 14 әҫәр, 476 MP3 файл, барлығы 78 сәғәт 40 минут. Эсенә шиғырҙар, романдар, повестар, әкиәттәр һәм башҡорт халыҡ эпостары (ҡобайырҙар) инә. Яҙмалар М. Х. Тухватшин исемендәге Башҡортостан республика һуҡырҙар өсөн махсус китапханаһы (БРСБС) тарафынан 1989–2019 йылдарҙа «һөйләүсе китап» рәүешендә әҙерләнгән.

Әҫәрҙәр дөйөм ҡулланылыштағы мираҫ (public domain) булып тора; датасет CDLA-Permissive-2.0 лицензияһы аҫтында таратыла.

Телдәр

Төп тел — башҡорт теле (ba). Бер әҫәр — «Урал батыр» эпосы — башҡорт һәм рус телдәрендә (ru) яңғырай. Һәр файлдың теле language бағанаһында күрһәтелә.

Ҡоролошо

Һәр әҫәр data/ папкаһындағы айырым папкала (латин хәрефтәре менән аталған); аудио өлөштәр уҡыу тәртибендә һанланған (0001.mp3, 0002.mp3, …). Оригиналь башҡортса һәм русса исемдәр metadata.csv файлында һаҡлана. Һәр яҙма тураһындағы мәғлүмәт (автор, уҡыусы, төр, йыл, оҙонлоҡ һ.б.) — шул уҡ metadata.csv файлында.

Ҡулланылышы

• Аҙ ресурслы төрки теле өсөн һөйләш моделдәрен алдан өйрәтеү (pretraining).
• Автоматик һөйләште таныу (ASR) — үҙ текстарығыҙҙы өҫтәгәндә.
• Текстан тауышҡа (TTS) һәм тауыш тикшеренеүҙәре.

⚠️ Был версияла текст-транскрипциялар юҡ — тик аудио ғына.

Лицензия һәм хоҡуҡтар

Датасет CDLA-Permissive-2.0 лицензияһы аҫтында (тулы тексты LICENSE файлында). Нигеҙендәге әҫәрҙәр — дөйөм ҡулланылыштағы мираҫ (РФ Граждан кодексы, 1282-се статья). Лицензия атрибуция талап итмәһә лә, сығанаҡты күрһәтеү хуплана. Атрибуция һәм сығанаҡ тураһында — NOTICE файлында.

Рәхмәт һүҙе

Тауыш яҙмалары — БРСБС (Өфө). Баҫтырыусы — Башҡорт телен һаҡлау һәм үҫтереү буйынса АНО.

---

Русский

Краткое описание

Датасет содержит аудиокниги преимущественно на башкирском языке: 14 произведений, 476 MP3-файлов, всего 78 ч 40 мин. В него входят стихи, романы, повести, сказки и башкирские народные эпосы (кубаиры). Записи подготовлены Башкирской республиканской специальной библиотекой для слепых им. М. Х. Тухватшина (БРСБС) в 1989–2019 годах как «говорящие книги» для незрячих читателей.

Литературные произведения находятся в общественном достоянии; датасет (компиляция записей и метаданных) публикуется под лицензией CDLA-Permissive-2.0.

Языки

Основной язык — башкирский (ba). Одно произведение — эпос «Урал-батыр» — звучит на башкирском и русском (ru) языках. Язык каждого файла указан в столбце language.

Структура

Каждое произведение — отдельная папка в data/ с латинским именем; аудиофрагменты пронумерованы в порядке воспроизведения (0001.mp3, 0002.mp3, …). Оригинальные башкирские и русские названия сохранены в metadata.csv, а не в путях к файлам. Полные сведения о каждой записи (автор, чтец, тип, год, длительность и т. д.) — в файле metadata.csv.

Применение

• Предобучение речевых моделей (self-supervised) для малоресурсного тюркского языка.
• Автоматическое распознавание речи (ASR) — при добавлении собственных транскрипций.
• Синтез речи (TTS) и речевые исследования.

⚠️ В этом релизе транскрипции не включены — только аудио.

Лицензия и права

Датасет — под лицензией CDLA-Permissive-2.0 (полный текст в файле LICENSE, распространяется вместе с данными согласно п. 2.1). Произведения — общественное достояние (п. 2 ст. 1282 ГК РФ). Лицензия не требует указания авторства, но ссылка на источник приветствуется. Атрибуция и происхождение — в файле NOTICE.

Благодарности

Записи — БРСБС (Уфа). Издатель — АНО по сохранению и развитию башкирского языка.