
Advances in Transformer-Based Text-to-Speech: Scaling with Knowledge Distillation
Spotify's research scales Transformer-based text-to-speech models using knowledge distillation, reducing size by 50% and doubling speed while improving quality.