Datasets | Mozilla Data Collective

This dataset consists of segmented Hausa speech audio clips paired with text, totalling 5h 25m and 38s.

Hausa-TTS-Dataset

License: NOODL-1.0

Locale: hau

Task: TTS

Format: MP3, TSV

Size: 276.90 MB

MDC DatasetsCompensated

🚧 Khowar-English Parallel Corpus 🚧

A parallel corpus of over 10,000 sentences of Khowar and English. The sentences are derived from a multimodal corpus consisting of 152 transcribed and translated Khowar language videos.

License: MDC DLA-1.0

Locale: khw, en

Task: MT

Format: TSV

Size: 407.57 KB

The dataset includes audio files in Persian and their translations in English.

CoVoST 2 Persian - English

License: CC-BY-NC-4.0

Locale: fa, en

Task: MT

Format: MP3, TSV

Size: 1.52 GB

Multimodal dataset for ɣɔmáláʔ (Ghomala, Bandjoun variety): IPA-annotated ritual corpus, 369 audio clips (≈17 min), and a Ghomala–French parallel corpus.

Ghomala-MultimodalDataset

License: NOODL-1.0

Locale: bbj

Task: NLP

Format: TSV, MP3

Size: 18.67 MB

Taruen

Chuvash TTS

A ~5-hour speech dataset for Chuvash Text-to-Speech (TTS) research, featuring a single female speaker reading news and digits at a rapid tempo.

License: CC-BY-SA-4.0

Locale: cv

Task: TTS

Format: PARQUET

Size: 854.02 MB

The dataset includes audio files in English and their translations in Persian.

CoVoST 2 English - Persian

License: CC-BY-NC-4.0

Locale: en, fa

Task: MT

Format: MP3, TSV

Size: 12.56 GB

A collection of read speech recordings in Kurmanji Kurdish.

Common Voice Scripted Speech 26.0 - Kurmanji Kurdish

License: CC0-1.0

Locale: kmr

Task: ASR

Format: MP3

Size: 1.92 GB

A collection of read speech recordings in Azerbaijani.

Common Voice Scripted Speech 26.0 - Azerbaijani

License: CC0-1.0

Locale: az

Task: ASR

Format: MP3

Size: 37.48 MB

Tupuri (tui) scripted speech: 2,085 clips (≈2h46min) from different speakers across two dialect subsets — Bango and Banwere.

Tupuri-ASR-Dataset

License: NOODL-1.0

Locale: tui

Task: ASR

Format: MP3,, TSV

Size: 97.82 MB

A collection of spontaneous responses to questions in Gorani.

Common Voice Spontaneous Speech 4.0 - Gorani

License: CC0-1.0

Locale: hac

Task: ASR

Format: MP3

Size: 225.34 MB

The dataset includes audio files in Turkish and their translations in English.

CoVoST 2 Turkish - English

License: CC-BY-NC-4.0

Locale: tr, en

Task: MT

Format: MP3, TSV

Size: 207.40 MB

The dataset includes audio files in English and their translations in Turkish.

CoVoST 2 English - Turkish

License: CC-BY-NC-4.0

Locale: en, tr

Task: MT

Format: MP3, TSV

Size: 12.56 GB

A collection of read speech recordings in Zaza.

Common Voice Scripted Speech 26.0 - Zaza

License: CC0-1.0

Locale: zza

Task: ASR

Format: MP3

Size: 54.88 MB

Kaleem Art Press

Saraiki-English Parallel Corpus

English–Saraiki parallel corpus: 51,447 aligned sentence pairs (~0.89M words), translated by Kaleem Art Press for MT and Saraiki NLP research.

License: CC-BY-NC-4.0

Locale: mul

Task: MT

Format: CSV

Size: 1.92 MB

A collection of spontaneous responses to questions in Bashkir.

Common Voice Spontaneous Speech 4.0 - Bashkir

License: CC0-1.0

Locale: ba

Task: ASR

Format: MP3

Size: 5.10 MB

MDC DatasetsCompensated

🚧 Moroccan Arabic ASR dataset (180h) 🚧

180 hours of fully segmented and transcribed Moroccan Arabic speech.

License: MDC DLA-1.0

Locale: ary

Task: ASR

Format: MP3, TSV

Size: 214 B

Multimodal dataset for Kekem (mbo): IPA-transcribed lexical datasheet, 277 audio clips (≈18m 47s) across 4 sessions, with audio alignment files.

Kekem-ALCAM-MultimodalDataset

License: NOODL-1.0

Locale: mbo-kekem

Task: NLP

Format: MP3, TSV

Size: 12.88 MB

Taraaz

Multilingual Humanitarian Response Eval (MHRE)

This multilingual humanitarian dataset contains 655 annotated datapoints evaluating AI chatbot safety and quality in migration and asylum scenarios across four language pairs (English–Farsi (Iranian Persian), Arabic, Kurdish (Sorani), Pashto). Built from 120 expert prompts, it includes outputs from GPT-4o, Gemini 2.5 Flash, and Mistral Small. The dataset provides both human evaluations from Respond Crisis Translation native-speaker evaluators and LLM-as-judge assessments (Gemini 2.5 Flash).

License: CC-BY-NC-SA-4.0

Locale: mul

Task: LLM

Format: csv

Size: 2.15 MB

A collection of read speech recordings in Uyghur.

Common Voice Scripted Speech 26.0 - Uyghur

License: CC0-1.0

Locale: ug

Task: ASR

Format: MP3

Size: 9.69 GB

A collection of read speech recordings in Uzbek.

Common Voice Scripted Speech 26.0 - Uzbek

License: CC0-1.0

Locale: uz

Task: ASR

Format: MP3

Size: 5.02 GB

Rerooted Archive

Rerooted: Speech Corpus of Testimonials from Armenian Refugees and Immigrants

Rerooted is an open-access online YouTube corpus of interviews with Armenian refugees and immigrants. As of now, the online corpus has over 80hrs of recordings on YouTube, alongside subtitles in Western Armenian. The current dataset is 20 hours from the corpus that we have curated with corrected and finely-annotated transcripts. The dataset is for use in NLP research, and we hope to continuously update the dataset with more curated transcripts.

License: GPL-3.0

Locale: hy; hyw

Task: ASR

Format: WAV, TEXTGRID

Size: 6.21 GB