wenet-e2e / wespeaker

Research and Production Oriented Speaker Verification, Recognition and Diarization Toolkit
Apache License 2.0
598 stars 104 forks source link

speaker embedding 提取问题请教 #279

Closed FengrunZhang closed 4 months ago

FengrunZhang commented 4 months ago

目前使用Voxceleb2训练好了Resnet34模型,在提取Voxceleb1的embedding时发现每次模型没有任何变化的情况下,对相同的句子每次提取出的embedding是有不同的,whole_utt已经设置为True了,请问还有哪些因素会导致上面的这种情况?