Dadaab refugee camp. Hundreds of thousands of Kenyan Somali speakers with near-zero ASR coverage. Pt 1 / 68-hour corpus just dropped, annotated in-region by @DigitalDivideData. The kind of dataset that actually changes something. 🎙️ #OpenData #ASR datacollective.mozillafoundation.org/datasets/cmm...

How do we stop LLMs from flattening culture? Join EM Lewis-Jong, CEO of @mozdatacollective.bsky.social, and Slow AI's Dr. Sam Illingworth live at 9am ET TODAY as they discuss why language and culture should exist in multitudes. open.substack.com/live-stream/...

Own your AI stack or rent it? @samillingworth.com speaks to @mozdatacollective.bsky.social on Substack Live substack.com/@samillingwo...

- the government’s mixed public messaging on AI & copyright is hindering licensing - the government should make a clear public statement that AI companies operating in the UK need to license their training data (which is the law) 4/5

They say: - the government must not weaken copyright law, and should instead strengthen licensing, transparency & enforcement - the government should stop prioritising large multinational tech firms 3/5

- AI training isn’t ‘learning’ and shouldn’t be treated as such The House of Lords has been absolutely consistent on this, and they are totally right. Will the government listen? /end

datacollective.mozillafoundation.org/datasets/cmm... Curated collection of public domain literature from Finland! The dataset captures the literary landscape of early 20th-century Finland and includes independent texts in both of the country's official languages: Finnish (fi) and Swedish (sv).

Discover new multilingual ASR, NLP, and TTS datasets on Mozilla Data Collective this week, and check out some site fixes and API improvements we've made! community.mozilladatacollective.com/mdc-release-...

1/5 Let's highlight one of our amazing text-to-speech contributors shaping AI data for African cultures. The Institute of African Digital Humanities has uploaded thousands of TTS audio clips totalling over 6 GB of data for more than 10 locales.

Community contributions play an enormous role in the development of multilingual AI. Here are 4 Javanese datasets to level up your TTS project with genuine, local dialects (all CC-BY-SA-4.0). Links in the comments!

The languages absent from your model are the communities it will fail to serve. Cultural preservation can be powered by technology. The Gojri (Gujari) Literature Corpus has over 60K tokens of clean, UTF-8 normalized text to make AI more accessible to Gujjar communities. Link in the comments!

LLM hallucination is a data problem as much as a model problem. Structured, machine-readable World Factbook corpus. Geopolitical, demographic & economic facts for every country – purpose-built for RAG grounding and knowledge graph construction. Last clean snapshot before the CIA retired the site.

📣 114 hours of recorded Nahuatl with transcripts! Get native speech from the municipalities of Zacatlán and Tepetzintla to take your NLP project to the next level. This is what responsible AI training data looks like. Explore the collection 👇 datacollective.mozillafoundation.org/datasets?q=n...

"📚 Over 18K proverbs, 243 tales & 380 legends! Tauren's Tatar Folklore helps tech get beyond multilingual by providing multiCULTURAL high-quality datasets. ⚖️ CC0, free for NLP projects: datacollective.mozillafoundation.org/datasets/cml...

@mozilla-fr.bsky.social Nouveaux jeux de données pour la traduction français–langues africaines! 🌍 datacollective.mozillafoundation.org/datasets/cml... datacollective.mozillafoundation.org/datasets/cml... datacollective.mozillafoundation.org/datasets/cmk...

🧡📣 1 million tokens of Balochi just arrived @mozdatacollective.bsky.social! Check out this beautiful dataset of journalistic Western Balochi (Rakhshani) datacollective.mozillafoundation.org/datasets/cml... @mozilla.org

Team @mozdatacollective.bsky.social is looking for paper submissions to support their session on Synthetic Speech at @interspeech.bsky.social find out if your work is a fit! interspeech2026.org/en-AU/pages/...

3 of our key staff including @kathyreid.au are delighted to be chairing a Special Session at @interspeech.bsky.social this year: Safeguarding Synthetic Speech: Ethical, legal and technical perspectives. The #CfP is now open and you can learn more at: safeguardingsyntheticspeech.org 1/3

NEW! Text-to-Speech for Tolaki from the amazing community in #Indonesia datacollective.mozillafoundation.org/datasets/cml...

Welcoming Mandar-Indonesian to Mozilla Data Collective! datacollective.mozillafoundation.org/datasets/cml...

We just released version 0.3.0 of our Python library with resumable downloads and other efficiency improvements. Now, downloading a dataset will now continue where a previous download left off - especially helpful for working with larger datasets. mozilla-data-collective.github.io/datacollecti...

1/7 🌍 The internet belongs to everyone - but AI still can't understand millions of African language speakers. That's a problem we can solve together! Here's what's now available on Mozilla Data Collective 🧵

Latest open data drop! 🔥 from our ASEAN language communities 🌏 read: community.mozilladatacollective.com/latest-asian... download: datacollective.mozillafoundation.org we can't wait to see what you build @mozilla.org @asianstudies.org @complingjournal.bsky.social @everythingopen.au #opensource 🧡

📣NEW! SmolTalk-Chinese: high-quality multi-task dataset for Chinese conversational AI, covering 19 task types: advice-seeking, code gen, daily chat! Dialogue model training & LLM fine-tuning 🧡 📎 Available at datacollective.mozillafoundation.org/datasets/cmk... ⚠️ Academic/non-commercial use only

📣🌍 Meet the Ewondo_Fong_ALCAM-MultimodalDataset! Highlighting Fong; a variant of Ewondo speech that’s widely used but rarely documented. Native-speaker audio, example sentences + Fong–French text to support language learning and speech tech 🧡 🔗 datacollective.mozillafoundation.org/datasets/cmk...

Thinking back to how many times we picked up a new information tool (and often put down an old one)… Also, announcing EM Lewis-Jong from @mozdatacollective.bsky.social as a speaker for #PfP✨London

📣 New dataset! 🇫🇷 Large-scale collection of 400,000+ real user conversations generated on the compar:IA platform. #opensource and proprietary systems developed within the French Ministry of Culture. 👏 datacollective.mozillafoundation.org/datasets/cmk... @info.gouv.fr @mozilla.org

Who gets to shape the tech we use? Since launching in Nov datacollective.mozillafoundation.org has been so excited about the educators, media orgs and technologists joining us! New in: • Teacher-led bias dataset (Argentina) • 70 hours of Luhya speech data (Kenya) • Radio archives (Northern Uganda)

Communities built the world’s largest open voice dataset. Now they’re building a new model for AI data — one rooted in choice, transparency, and collective power. Mozilla Data Collective is live🎉 ✳️ Join the community mzl.la/44aTEoe @mozdatacollective.bsky.social #MozillaDataCollective

Will you be in Melbs for @pyconau.bsky.social #PyConAU #PyConAU2025 in the next few days? Don't miss MDC's @kathyreid.au speaking on Sun 14th in Ballroom 1 at 1410hrs - introducing a better way to provide and source #data for #AI. Sneak peek: community.mozilladatacollective.com/your-dataset...

Hello world! We're the Mozilla Data Collective - and we want to rebuild the #AI #data #ecosystem - with #communities at the centre. 🚫 Unconsented web scraping 🚫 Sign license deal with megacorp ✅ The Mozilla Data Collective Launching Wednesday! 🚀 💫 datacollective.mozillafoundation.org