rtzr/Awesome-Korean-Speech-Recognition

Benchmarks character error rates (CER) across six major Korean speech recognition APIs—including OpenAI Whisper, Google Cloud, ETRI, Naver Clova, and others—against six AI-Hub datasets covering diverse acoustic conditions (meetings, counseling, low-quality telephony, lectures). Uses normalized CER rather than WER to account for Korean morphosyntactic complexity, providing standardized evaluation methodology and metrics libraries for character-level assessment.

492 stars. No commits in the last 6 months.