based on: Self-attention encoding and pooling for speaker recognition reference: https://github.com/Jungjee/RawNet https://github.com/kaituoxu/Speech-Transformer https://wikidocs.net/31379