Hi Lirry, excited that you’re far along in your work using Spokestack!
To create a model with robust recall across different speakers we’ve found that roughly 1,000 different people with at least three samples each is needed.
That’s a lot, I know! I will contact you about participating in the beta of our new automated data collection service that addresses exactly this problem.
Since Maker is targeted at individual use and prototyping, it does not have that capability. But we think the automated data collection feature for Pro that is now in beta will let you accomplish the same thing more easily. If you already have a robust set of samples you’d like to use for model training, let me know and we can work out a suitable Enterprise plan with BYO data included!