We developed a new ML model, and as a result, the training time become x4K faster than previous LSTM, and the accuracy is the same or better. This link is the demo of detecting speakers using our model.
The API and pip installation will be coming soon.