Emilia-large and SeamlessAlign
#2
by HarryHe - opened
Nice work and exciting resources! Just a quick note: Emilia-Large (open-sourced a year ago, 216k hours, https://arxiv.org/abs/2501.15907) and SeamlessAlign (270k hours) are both larger, so ‘largest ever’ may be a slight overstatement.
Thanks for pointing this out — and for the reference. Our claim specifically refers to being the largest open-source multilingual speech dataset with word-level alignment. Emilia-Large and SeamlessAlign are excellent and complementary resources in terms of scale.
HarryHe changed discussion status to closed