18.97.14.81
18.97.14.81
close menu
Accredited
한국어 음성 언어 처리의 두 문제 - 맥락 의존성과 음향 지각의 불분명성을 중심으로 -
Two issues in Korean speech language processing - Context dependency and acoustic ambiguity perception -
김아름 ( Kim A-reum )
UCI I410-ECN-0102-2023-000-001096084

This study aims to examine how theoretical problems in speech recognition models manifest as recognition errors in actual speech recognition situations, with a particular focus on the issues of context dependency and acoustic ambiguity. To this end, we conducted 376,830 seconds of speech recognition experiments and discussed how context dependency in speech recognition models leads to errors when contextual information becomes unstable due to homophonic forms, decreases recognition performance for low-frequency words and proper nouns, and causes overcorrection problems in phonemic restoration based on context. We also argued that acoustic ambiguity results in recognition errors between final consonants and in the speaker identification domain based on individual acoustic characteristics, leading to numerous recognition errors.

1. 서론
2. 맥락 의존성에 의한 음성 인식 오류
3. 음향 지각의 불분명성에 의한 음성 인식 오류
4. 결론
[자료제공 : 네이버학술정보]
×