CLOVA Speech: End-to-End Speech Recognition in Everyday Life -English version- скачать в хорошем качестве

CLOVA Speech: End-to-End Speech Recognition in Everyday Life -English version-

Hyuksu Ryu / LINE ■ Session Overview The conventional paradigm in automatic speech recognition (ASR) has been based on several independently trained modules such as an acoustic model which predicts acoustic features given phoneme sequences, and a language model which predicts probability between words, given the acoustic features. However, there has been a growing amount of attention on end-to-end (E2E) speech recognition, which goes one step beyond traditional speech recognition to predict word sequences directly from the input speech using a simple neural network. Particularly, in recent times, end-to-end speech recognition functions have finished the textual experimental phase and are gradually becoming a crucial axis for actual commercialized services. This presentation will focus on current and planned service cases for our, currently under development, multilingual E2E speech recognizer (NEST) in Japanese, Korean, and English. ■ Official Site #linedevday https://linedevday.linecorp.com/2020/... ■ Category AI&Data - AI