How many wake-up words can be supported simultaneously at most？

[SeniorGlassMaster]

Ignore model size, computation latency, and computing resources, etc？

[mlxu995]

It depends on your purpose. In the case of sufficient resources, it is feasible for any limited number of keywords. But if you need to support unplanned keywords, the best way is using CTC decoding.