In this work, we adopt text-to-speech data augmentation to improve the performance of children’s speech recognition system. We find that the children’s TTS model generates speech with inconsistent ...