Sure it definitely had some strange moment but it’s impressive for a small training data set and low compute
That doesn't sound like the same bar as "most authentic full-duplex speech model ever built"
That doesn't sound like the same bar as "most authentic full-duplex speech model ever built"