Kimi-Audio Technical Report

Benchmark Model Rank Results
audio-question-answering-on-openaudiobench-llamaquestionsKimi-Audio#3Accuracy: 79.33
speech-recognition-on-librispeech-test-otherKimi-Audio#7Word Error Rate (WER): 2.42
speech-recognition-on-librispeech-test-cleanKimi-Audio#19Word Error Rate (WER): 1.28
audio-understanding-reasoning-on-mmauKimi-Audio (8.2B)–Accuracy (Avg, Test-mini): 68.20