5fa1a76
1
2
3
This model has four main components: A feature encoder takes the raw audio waveform, normalizes it to zero mean and unit variance, and converts it into a sequence of feature vectors that are each 20ms long.
