@nihilists4jesus Same thing with 50cc mopeds in the 70s/80s. They were a legal gray area and you could ride them without a license and underage. Not so much anymore!
This week we launched Bland Speech v3: the world's most realistic TTS model, outranking ElevenLabs, Cartesia, OpenAI, and xAI.
Today, our research team sat down to discuss how we built the model, the unique approach we took, and the impact it'll have.
It's official โ Bland Speech v3 is the world's most realistic voice model according to @designarena
Hear what Matthew and Karen have to say on the topic ๐
Try it yourself for free at https://t.co/E8Hbzk2gnp
Bland researcher Pedro Sandoval spent his PhD figuring out how to break AI models.
Today, he helps build the world's first Human Speech Engine โ the only text-to-speech model trained entirely on real conversations.
Take a listen to what he has to say about how Bland's research team built a model that outperforms every other frontier lab on realism benchmarks.
@danc_danc@LinusEkenstam I've found that it's a little voice-dependent. Some of them handle Slavic languages or Spanish better than others, for example.
@benyuls@heyfredds Like anything, the cleaner, the better. The audio we got from James's videos was pretty rough to begin with, but the results speak for themselves.
(source: I work at Bland)