Reproduce the test results of xttsv2

[yeyubeifeng]

Why are there so many differences between the test results I reproduced for xttsv2 and the results presented in the paper.The text dataset used is the FLORES+ dataset from the paper, and the reference audio is the corresponding reference audio from ZS-TTS-Evaluation.. For XTTS inference I have used a temperature equal to 0.75, length penalty equal to 1.0, repetition penalty
equal to 10.0, top k equal to 50, and top p equal to 0.85.