표 4. | Table 4. Command-line으로 inference한 Whisper 모델의 크기별 음절 오류율(CER, %)[libri-other의 경우, 단어 오류율(WER, %)로 측정됨] | CER for each size of the Whisper model inferred by command-line (for libri-other, measured by WER)

테스트 데이터셋	large-v2	medium	base	tiny
kspon-evalclean	10.83	11.94	22.85	33.57
kspon-evalother	10.83	11.96	21.12	30.90
spon-bcast	13.14	14.03	21.30	28.49
spon-debate	10.25	11.48	18.65	25.97
spon-present	11.14	11.49	16.76	21.91
libri-testother	6.31	6.74	11.95	15.79

CER, character error rate; WER, word error rate.