The traditional approach to speech recognition with several intermediate parts is exemplified as follows:

$Audio\:\left(input\right)->feature\:extraction->phoneme\:detection->word\:composition->text\:transcript\:\left(output\right)$

The end-to-end approach replaces the intermediate chain into a single deep neural network, allowing the use of a single optimization criterion for enhancing the system:

$Audio\:\left(input\right)-\left(Neural\:Network\right)->text\:transcript\:\left(output\right)$