Datasets | Mozilla Data Collective

This dataset presents 21 video-recorded Ekpeye folktales (1h28m) narrated by two community elders, each paired with transcripts and English translations that include narrative summaries. It offers a rich multimodal resource for speech, video, storytelling, and cultural heritage research, as well as training multilingual and multimodal AI systems.

Documenting Ekpeye Folktales and Preserving Cultural Heritage

License: CC-BY-NC-SA-4.0

Locale: ekp

Task: OTH

Format: MP4, TXT, DOCX

Size: 5.97 GB

This dataset contains 98 recordings (≈1.16 hours) of everyday Tyap speech from 10 community speakers, each paired with detailed transcripts and English translations.

Atyap Afwan_: Preserving Tyap Through Community-Driven Speech Data

License: CC-BY-NC-SA-4.0

Locale: kcg

Task: NLP

Format: WAV, TXT

Size: 251.51 MB

Institute of African Digital Humanities

Basaa-ALCAM-MultimodalDataset

This dataset comprises a datasheet of lexical entries in Basaa, accompanied by illustrative sentences, word-by-word glosses, and corresponding translations in French. Each entry is enriched with aligned audio recordings, making the resource suitable for linguistic analysis and speech technology development.

License: NOODL-1.0

Locale: bas

Task: NLP

Format: MP3, TSV

Size: 14.66 MB

Common Voice

Mozilla Common Voice Spontaneous Speech ASR Shared Task Test Data

A bundle of the held-out test data for the Mozilla Common Voice Spontaneous Speech ASR shared task.

License: CC0-1.0

Locale: mul

Task: ASR

Format: MP3, TSV

Size: 784.80 MB

This dataset offers 11.3 hours of natural everyday speech in Ibọnọ and Obolo, captured from 20 adult speakers across 120 recordings, each paired with a clean transcript and metadata.

Everyday Interactions in Ibọnọ and Obolo Languages

License: CC-BY-NC-SA-4.0

Locale: ibn, ann

Task: NLP

Format: WAV, TXT

Size: 2.43 GB

TidyVoice2026 Challenge

TidyVoiceX_ASV

This dataset is designed for speaker verification using the Mozilla Common Voice corpus across 40 languages. It includes approximately 5,000 speakers who each have recordings in more than one language. Leveraging this multilingual overlap, we construct the trial pairs to explore cross-lingual variation in the speaker verification task.

License: CC0-1.0

Locale: mul

Task: OTH

Format: WAV

Size: 36.72 GB

This dataset contains audio recordings of Bible verses in Ehugbo, a dialect of Igbo (a Niger-Congo language spoken in Nigeria). It contains 312 audio recordings of biblical text-to-speech data comprising 1 hour and 30 seconds of speech data.

Ehugbo TTS: biblical text to speech dataset in Ehugbo Language

License: CC-BY-NC-SA-4.0

Locale: ig-ehugbo

Task: TTS

Format: WAV

Size: 437.69 MB

This dataset contains audio recordings of the Nupe language. It features 1,583 audio recordings comprising 2 hours, 40 minutes, and 32 seconds of speech data, with paired transcripts. The recordings feature 8 unique speakers representing three distinct Nupe accent varieties: Bida accent, Kutigi accent, and Lapai accent.

Speech Data Collection for The Nupe Language

License: CC-BY-NC-SA-4.0

Locale: nup

Task: NLP

Format: WAV, TXT

Size: 1.58 GB

Akylai

KyrgyzNER: Human-Annotated NER Dataset for Kyrgyz

KyrgyzNER is the first manually annotated Named-Entity Recognition (NER) dataset for the Kyrgyz language. It consists of 1,499 news articles (10,900 sentences, 140k tokens) from the 24.kg news portal, annotated with 39,075 entity mentions across 27 classes.

License: CC-BY-NC-SA-4.0

Locale: ky

Task: NLP

Format: CONLL-2003

Size: 585.87 KB

1.3 million tokens in the Putèr variety of Romansh from the daily newspaper “La Quotidiana”.

Putèr Newspaper Corpus

License: CC0-1.0

Locale: rm-puter

Task: OTH

Format: TSV

Size: 8.94 MB

1.3 million tokens in the Sutsilvan variety of Romansh from the daily newspaper “La Quotidiana”.

Sutsilvan Newspaper Corpus

License: CC0-1.0

Locale: rm-sutsilv

Task: OTH

Format: TSV

Size: 8.87 MB

14.6 million tokens in the Sursilvan variety of Romansh from the daily newspaper “La Quotidiana”.

Sursilvan Newspaper Corpus

License: CC0-1.0

Locale: rm-sursilv

Task: OTH

Format: TSV

Size: 37.80 MB

6.1 million tokens in the Rumantsch Grischun variety of Romansh from the daily newspaper “La Quotidiana”.

Rumantsch Grischun Newspaper Corpus

License: CC0-1.0

Locale: rm-rumgr

Task: OTH

Format: TSV

Size: 19.03 MB

Community

Podcast Homostoria (Indonesia)

This dataset features discussions on modern media—including film, podcasts, and social media—and its connection to local customs and traditions. The conversations are primarily in Indonesian, with frequent code-switching between English and Javanese.

License: CC-BY-SA-4.0

Locale: id

Task: ASR

Format: mp3

Size: 302.97 MB

Text to speech dataset for Hungarian, male speaker, approximately 1.5 hours of read speech.

Imre 1.0

License: CC0-1.0

Locale: hu-HU

Task: TTS

Format: WEBM

Size: 99.60 MB

Text to speech dataset for Hungarian, female speaker, approximately 1 hour of read speech.

Berta 1.0

License: CC0-1.0

Locale: hu-HU

Task: TTS

Format: FLAC

Size: 209.52 MB

Text to speech dataset for Hungarian, female speaker, approximately 1.5 hours of read speech.

Anna 1.0

License: CC0-1.0

Locale: hu-HU

Task: TTS

Format: WEBM

Size: 95.27 MB

Text to speech dataset for Spanish, male speaker, approximately 1.5 hours of read speech.

Dave 1.0

License: CC0-1.0

Locale: es-ES

Task: TTS

Format: WEBM

Size: 85.24 MB

Text to speech dataset for English, female speaker, approximately 1 hour of read speech.

Kathleen 1.0

License: CC0-1.0

Locale: en-US

Task: TTS

Format: FLAC

Size: 211.96 MB

Text to speech dataset for English, male speaker, approximately 1 hour of read speech.

Joe 1.0

License: CC0-1.0

Locale: en-US

Task: TTS

Format: WEBM

Size: 75.78 MB