Povey's law

Povey's law is an observation from speech recognition research that suggests the number of computational resources needed—like the processing power and data—grows roughly proportionally to the square of the complexity involved in understanding speech. Specifically, as speech models become more detailed or accurate, the effort to train and run them increases significantly, roughly following a quadratic relationship. This means that improving speech recognition systems often requires disproportionately more data and computing capacity, highlighting the challenge of achieving higher accuracy in speech technology as models grow more sophisticated.