End-to-end (E2E) models fold the acoustic, pronunciation and language models
of a conventional speech recognition model into one neural network with a much
smaller number of parameters than a conventional ASR system, thus making it
suitable for on-device applications. For example, recu