Generated mel is shorter and speech is too fast

[soyplane]

Hi，
Thanks for your work! I tried to reproduce your results on LibriSpeech and found that the generated speech is too fast. Currently, it appears that the generated mel-spectrograms are generally shorter compared to yours. Have you encountered a similar situation and attempted to resolve it? Below are my experimental configuration and final results.