2 ms·

can any of these models be coerced into just doing straight phonetic transcription? like spitting out IPA?