What does batchsize mean in AAAI 19 presentation

[tu1258]

I found that random sparsity and balanced sparsity has comparable inference speed when batchsize=1. But there is only 1 audio when decoding in real world, so the batch size is always 1, right?
Correct me if I mislead, thanks!