automatic transcription is a system to change phonemes to graphemes automatically, it's more like syllabel recognition, used for build speech recognition in another language base on stable and existing version of speech recognition.
see this picture to more understanding: see this picture
one think that you must understood, sound is different with meaning, computer need tools to understood the sound, it's speech or music or another form of sound. all speech sound already mapped in International Phonetic Alphabet (IPA).
and with little computation to combine the graphemes (call dictionary), resulting understanding speech of specific language.
example: you already know the sound of "spoon" or 'key', the grapheme is split by s and 'p-o-on'.
in other language like Indonesia's language, another developers use this grapheme identification system (transcription system) to build a speech recognition, for words 'meskipun'.
in grapheme, 'meskipun' words is: 'mehs-key-poon'.
with a simple computation, we can make computer understood the 'meskipun' words.
only said if "mesh + key + poon" show, the words is 'meskipun'
meskipun (=althought, english)
The big problem in building speech recognition with using automated transcription system, there is no 100% similar for IPA map in every language.
so the developers should use several 'transfer learning' database of language to make their speech recognition have more high accuracy. except, he decided to build it from scratch.
automated speech recognition is an end to end speech recognition that set to understood a specific speech language, which in the middle of the system contain automatic transcription system.