@ostrisai if you try to look at the results of recent benchmarks (like https://t.co/GAFPDlZuOm), Qwen3 and Internvl3.5 families come to mind. Comparison in the article on Muse Glimmer 30b also points to this.
@sievedata Hey @sievedata, it's a great effort to unify different parts of speech processing in one benchmark. What about lipsync algorithms, did they participate somehow in the assessment, or videos on the benchmark webpage is only for the clarity?
@skalskip92@danylo_movchan After your reply i tried again and got it ( i guess it's somehow connected to auth via kaggle or maybe they changed something).