Problem of cosine similarity computing for speaker verification

[Winterspringkle]

python verification.py --model_name ecapa_tdnn --wav1 vox1_data/David_Faustino/hn8GyCJIfLM_0000012.wav --wav2 vox1_data/Josh_Gad/HXUqYaOwrxA_0000015.wav
2023-08-03 18:48:40 | INFO | fairseq.tasks.text_to_speech | Please install tensorboardX: pip install tensorboardX
The similarity score between two audios is 0.9880 (-1.0, 1.0).

When I tried the script provided in readme, I got the similarity of 0.9880, which is far from the 0.2053 in readme and no help to speaker verification task. How to reproduce the result in readme file?

[RegulusBai]

Did you set the checkpoint?