Topics
late
AI
Amazon
Image Credits:MicroStockHub(opens in a new window)/ Getty Images
Apps
Biotech & Health
clime
Image Credits:MicroStockHub(opens in a new window)/ Getty Images
Cloud Computing
Commerce
Crypto
Enterprise
EVs
Fintech
Fundraising
Gadgets
back
Government & Policy
Hardware
layoff
Media & Entertainment
Meta
Microsoft
Privacy
Robotics
Security
Social
outer space
Startups
TikTok
Transportation
Venture
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
MLCommons , a nonprofit AI safety workings chemical group , has teamed up with AI dev platform Hugging Face to release one of the world ’s largest collections of public domain voice recordings for AI enquiry .
The dataset , calledUnsupervised People ’s Speech , hold more than a million hour of audio frequency spanning at least 89 languages . MLCommons order it was propel to create it by a desire to support R&D in “ various areas of speech technology . ”
“ Supporting encompassing natural lyric processing research for languages other than English helps bring communicating technology to more people globally , ” the system wrote in ablog postThursday . “ We previse several avenues for the research community to continue to build and develop , especially in the areas of better gloomy - resourcefulness language speech models , enhanced speech recognition across different accents and dialects , and refreshing applications in speech synthetic thinking . ”
It ’s an admirable end , to be sure . But AI datasets like Unsupervised People ’s Speech can deport risk for the researchers who choose to use them .
one-sided data is one of those risks . The transcription in Unsupervised People ’s Speech came from Archive.org , the non-profit-making perhaps best known for the Wayback Machine web archival tool . Because many of Archive.org ’s contributors are English - speaking — and American — almost all of the recordings in Unsupervised People ’s Speech are in American - accent English , per the readme on the official project pageboy .
That means that , without heedful filtering , AI system like actor’s line recognition and voice synthesizer model train on Unsupervised People ’s Speech could exhibit some of the same prejudice . They might , for example , struggle to transliterate English talk by a non - native utterer , or have trouble generating synthetic voices in languages other than English .
Unsupervised People ’s Speech might also contain recordings from people incognizant that their voices are being used for AI research purposes — admit commercial applications . While MLCommons aver that all recording in the dataset are public domain or useable under Creative Commons licenses , there ’s the possibility error were made .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
According to an MIT psychoanalysis , hundreds of publicly available AI training datasets lack licensing selective information and contain errors . Creator advocates including Ed Newton - Rex , the CEO of AI ethics - pore nonprofit Fairly Trained , have made the cause that creators should n’t be demand to “ prefer out ” of AI datasets because of the burdensome burden prefer out imposes on these creators .
“ Many Maker ( for instance Squarespace users ) have no meaningful way of opting out,”Newton - Rex wrotein a post on X last June . “ For creators whocanopt out , there are multiple overlap opt - out method acting , which are ( 1 ) incredibly confusing and ( 2 ) woefully incomplete in their coverage . Even if a perfect universal opt - out subsist , it would be hugely unfair to put the opt - out loading on creators , given that generative AI uses their piece of work to vie with them — many would simply not realise they could opt out . ”
MLCommons says that it ’s committed to updating , maintain , and improving the quality of Unsupervised People ’s Speech . But give the potential flaws , it ’d behoove developer to work out serious precaution .
