Experiments in OpenCLIP (main branch) with `timm` NaFlexViT audio encoder + modern text encoder for variable-time , variable-length text CLAP models