Model param num

[hajkeoadf]

Hello! Thank you for your great work!

I'm wondering if there is an estimation of the whole model parameters?

[dscripka]

Thanks, I'm glad you are finding the library useful!

The total model parameters are split between the Google speech embedding model (around ~400k parameters) and the trained wakeword model (typically <100k) parameters. There is some flexibility in how large the wakeword model can be, but performance only goes up a small amount with increasing parameters, as typically the quality of the training data is the limiting factor.