Datasets | Mozilla Data Collective

An Urdu multi-speaker TTS dataset distributed in 36 zip files, each containing audio files and a TSV mapping file, with approximately 10 hours of speech.

Urdu Multi-Speaker TTS Dataset

License: CC-BY-NC-4.0

Locale: urd

Task: TTS

Format: WEBM, TSV

Size: 514.54 MB

A large-scale Sindhi Part-of-Speech (SiPOS) tagging dataset for developing and evaluating NLP models in a low-resource language.

Sindhi Parts Of Speech (SiPOS)

License: CC-BY-NC-4.0

Locale: snd

Task: NLP

Format: TXT

Size: 874.04 KB

Curated Urdu dataset of 100,000 records with fabricated PII spans, for training and evaluating privacy span detection across 12 domains.

PII Redact Urdu UOPF

License: CC-BY-NC-SA-4.0

Locale: urd

Task: NLP

Format: JSONL

Size: 5.77 MB

A literature corpus of Wakhi language texts, preserving the language’s literary, cultural, and linguistic heritage for research and low-resource NLP.

Wakhi Literature Corpus

License: CC-BY-ND-4.0

Locale: wbl

Task: NLP

Format: DOCX, TXT

Size: 74.83 KB

Kaleem Art Press

Multilingual Religious Parallel Corpus (Kaleem Art Press)

This dataset is a multilingual parallel sentences corpus containing 6,465 aligned sentence units with approximately 0.98 million words, curated from Kaleem Art Press archives. It includes parallel religious text data in Arabic, Urdu, Saraiki (standard and dialectal), Punjabi (Shahmukhi), and English, supporting research in machine translation, comparative linguistics, digital humanities, and low-resource language studies.

License: CC-BY-SA-4.0

Locale: mul

Task: MT

Format: CSV

Size: 2.27 MB

Hindi monolingual corpus of about 10 million tokens, compiled from multiple authors and including literary works and personal articles.

Hindi 10 Million Text Corpus

License: CC-BY-ND-4.0

Locale: hin

Task: NLP

Format: DOCX

Size: 21.91 MB

A bilingual Urdu and Sindhi grapheme-to-phoneme dataset for G2P, TTS, ASR, and speech technology research.

Urdu-Sindhi Grapheme-to-Phoneme Dataset

License: CC-BY-NC-SA-4.0

Locale: ur, sd

Task: TTS

Format: PARQUET

Size: 1.92 MB

SdCorpus is a large scale monolingual Sindhi text corpus collected from publicly available web sources for NLP and language model development.

Sindhi Corpus

License: CC-BY-NC-4.0

Locale: snd

Task: LM

Format: TXT

Size: 321.00 MB

MDC Curators

CoVoST 2 Arabic-English

The dataset includes 5736 audio files in Arabic and their translations in English.

License: CC-BY-NC-4.0

Locale: ar, en

Task: MT

Format: MP3, TSV

Size: 148.21 MB

Urdu translation of Anthropic's HH-RLHF preference dataset for RLHF/reward-model research in Urdu.

Helpful and Harmless RLHF Urdu

License: CC-BY-NC-SA-4.0

Locale: urd

Task: NLP

Format: Parquet

Size: 10.37 MB

A 500K token corpus of Rangpuri literature and drama provided in UTF-8 format for linguistic research and NLP development.

Rangpuri (অংপুরি Ôṅgpuri) Text Corpus

License: CC-BY-NC-4.0

Locale: rkt

Task: NLP

Format: TXT

Size: 3.68 MB

A curated Marathi text corpus containing diverse written content for natural language processing, linguistic research, and language modeling tasks.

Marathi Blog & Literature Corpus

License: CC-BY-NC-4.0

Locale: mar

Task: NLP

Format: TXT

Size: 4.20 MB

Ormuri Literature Corpus is a collection of Ormuri literary texts by Rozi Burki, supporting research and preservation of this under-resourced language.

Ormuri Literature corpus by Rozi Burki

License: CC-BY-NC-SA-4.0

Locale: oru

Task: NLP

Format: PDF, DOCX

Size: 18.92 MB

Maseno Centre for Applied Artificial Intelligence (MCAAI)

DhoNam: Dholuo Speech dataset

DhoNam: Dholuo Speech dataset is a speech corpus designed to supercharge Automatic Speech Recognition (ASR) and other speech technologies for Dholuo, one of Kenya’s major indigenous languages. This dataset contains native-speaker audio recordings collected through a platform where users read aloud a displayed sentence. The dataset includes the audio recordings and the corresponding prompt/sentence that was read.

License: NOODL-1.0

Locale: Luo

Task: ASR

Format: WEBM

Size: 2.49 GB

Kaler Kontho

Kaler Kantho Bengali Newspaper Corpus

A 10+ million token Bengali newspaper corpus from Kaler Kantho provided in .docx format for large-scale NLP research and linguistic analysis.

License: CC-BY-NC-4.0

Locale: ben

Task: NLP

Format: DOCX

Size: 33.11 MB

CLEAR Global

TWB Parallel Sentence kits - Rohingya (5k)

5,000 English–Rohingya parallel sentences from CLEAR Global's Gamayun mini-kit, translated by professional translators from Tatoeba source sentences.

License: CC-BY-4.0

Locale: rhg

Task: MT

Format: TSV

Size: 358.88 KB

CLEAR Global

TWB Parallel Sentence kits - Kanuri (5k)

5,000 English–Kanuri parallel sentences from CLEAR Global's Gamayun mini-kit, translated by professional translators from Tatoeba source sentences.

License: CC-BY-4.0

Locale: kau

Task: MT

Format: TSV

Size: 358.46 KB

Information and Language Processing Research Lab

Nwāchā Munā

A 5.39-hour manually transcribed Devanagari speech corpus for Nepal Bhasha, with an ASR benchmark for low-resource speech research.

License: CC-BY-NC-4.0

Locale: new

Task: ASR

Format: WAV, TSV

Size: 477.62 MB

Kaleem Art Press

Kaleem Art Press Saraiki Literature Corpus

This corpus contains approximately one million tokens of Saraiki text curated over the past ten years by Kaleem Art Press. It features a wide range of literary genres, including stories, short stories, novels, fiction, non-fiction, travelogues, poetry, biographies, and historical writings. All content is shared with full author approval. The dataset is intended to support linguistic research, Saraiki language technology development, and the preservation of cultural and literary heritage.

License: CC-BY-NC-4.0

Locale: skr

Task: OTH

Format: TXT

Size: 1.84 MB

A 10.4-million-word Saraiki news corpus from Associated Press of Pakistan, covering 2016–2026 for non-commercial Saraiki NLP research.

Associated Press Pakistan 10.4 Million Saraiki Corpus

License: CC-BY-NC-SA-4.0

Locale: skr

Task: NLP

Format: TXT

Size: 19.51 MB