Record your voice. Earn $SLC + USDC. Teach AI your language. The dataset you can’t scrape, already training frontier AI models. Join 2.5M+ in 180 countries 🌍
USA
Joined October 2022
- Tweets2.8K
- Following197
- Followers153K
- Likes8.8K
Silencio | Voice Data for AI retweeted
In case you still don't get how huge Silencio could be, then you might have missed this:
Like a month ago, the Gates Foundation pulled together a 60-member coalition, Anthropic, Google, OpenAI, Microsoft, NVIDIA, ElevenLabs and many more, with one shared 5-year goal:
give 3.4 billion people access to AI in their own language and voice. that means closing the gap across more than 7,000 languages, most of which barely have the voice data to train on.
in other words, the biggest institutions and leading AI labs on earth have confirmed that voice data for underrepresented languages is one of the most important problems in AI right now.
that's the exact market @silencioNetwork has been building for this whole time, and likely no other firm in the world can serve niche languages from 2.5M+ people across the globe the way they can.
given the timing and relevance, maybe the most exciting DePIN in the entire space right now. and the fact that they're already working with some of the leading AI labs as paying clients validates exactly that.
Under the coordination of the Blocksound Foundation (@FoundationSLC), SLC is expanding to Solana.
Contract Address: 6aqQf9dsYAUwyW3jRawBBxZ1oy8ehmBYb519MhjV2eab
Solscan:
solscan.io/token/6aqQf9dsYAU…
Jupiter:
jup.ag/?sell=EPjFWdd5AufqSSq…
Meteora pool:
meteora.ag/dlmm/FtMJoFr6Shjn…
⚠️ Always verify the contract address (CA) posted on this account and silencio.network.
(Jupiter verification is in progress, and Solscan is still indexing our logo and metadata. Until that’s done, the token may show as unverified or without a logo.)
Powered by peaqOS from @peaq. The only official SLC on Solana.
This week in Munich at the Conversational AI & Customer Experience Summit.
Every voice agent in the room hits the same wall: it only understands people it has heard. For most of the world's languages, that audio does not exist yet.
Some of these companies already build on it.
Monthly Raffle is upon us!
The final 4 words are in.
(Same Procedure as every month)
Each vote count is turned into a number and combined into one string and hashed with SHA-256 to generate a provably random seed. That seed selects the winners.
Vote now 📷
34%Accent
28%Dialect
23%Acoustic
15%Dawn
364 votes • Final resultsFilipino-accented English. One public benchmark. Leading ASR systems: 10.8% to 44.7% word error rate.
Same accent, same test.
Choose your model on the accents your users have. What we heard at Interspeech 2026: silencio.network/updates
Before you trust a headline word error rate, ask two questions.
Who was speaking?
Were they reading?
Then build a test set that sounds like your users. silencio.network/updates
Today is Uzbek Language Day.
33 million speakers across Central Asia. Barely represented in public speech data.
If you speak Uzbek, yours is one of the voices AI has hardly heard. Record a clip. Change that.
silencio.network
Silencio is proud to support @Blockworks comment letter to the SEC on Regulation Crypto Assets.
Who recorded our Kenyan Swahili sample?
103 speakers, one clip each. 66 from Nairobi, where Sheng runs through everyday Swahili. 26 from Mombasa, on the coast.
Unscripted answers, recorded on their own devices, every clip human-transcribed.
huggingface.co/datasets/Sile…
Myth: AI understands every language.
Around 7,000 languages exist. Strong speech models handle a small fraction of them at production quality.
The gap is not technology. It is whether anyone recorded the language in the first place.
Every clip in our catalogue is recorded by a real person in their own environment: homes, streets and workplaces from Lagos to the Visayas to Addis Ababa.
That is the acoustic world voice AI is deployed into.
It is rarely the one it is tested on.
What language or dialect in your family is at risk of being left out of AI forever?
Reply. We want to hear which ones people are worried about.
Code-switching is where voice assistants break.
Hindi to English and back, mid-sentence. People do it without noticing. Models trained on one language at a time notice every time.
It is almost never read aloud, so it has to be recorded the way people talk. We collect it to spec.
Our speech data is recorded by ~2.5M contributors across ~180 countries, on their own phones, where they live and work.
Each one is paid per recording, with a consent record for every speaker.
That is why the accents in it are the accents real users have.
Your Voice is the missing Data.
Check campaigns available for you: ai.silencio.store/
How do you say "hello" in your language?