3D-Speaker

3D-Speaker copied to clipboard

小白请教一下这段代码

3

https://github.com/modelscope/3D-Speaker/blob/main/speakerlab/models/campplus/layers.py#L109 Does it imply that seg.shape[-1] should be less than x.shape[-1]? If not so, https://github.com/modelscope/3D-Speaker/blob/main/speakerlab/models/campplus/layers.py#L98 will certainly raise an error. Is this the underlying relationship between x and seg?

Juelianqvq

按照脚本预处理自有数据集，但发现内存爆炸，不足以支持pin_memory。 然后在config里关闭了pin_memory，但发现内存泄露。 按照我查看dataset和dataloader，按道理音频是在call的时候 才被读取，不应占有过多内存。从最初的1.2T慢慢增长到1.6T 请问有遇到过吗？

jinhonglu

I try to Training the SDPN network

7

I try to Training the SDPN network,and i use this page to train the sv-sdpn: https://github.com/modelscope/3D-Speaker/tree/main/egs/voxceleb/sv-sdpn and I got the following error message, please help me to fix it Stage2:...

kevin00616

VoxCeleb2训练集训练CAM++模型收敛速度与模型ERR，与仓库中所提出的结果差距甚大，与一般的说话人识别，在voxceleb2训练结果也差距甚大，是库版本或者训练机器还是其他什么之类的错误吗？

4

您好，在使用原来的python egs/voxceleb/sv-cam++/speakerlab/bin/train.py训练代码和原始的配置文件进行训练，，未修改代码，训练收敛速度和训练性能与一般的VoxCeleb2-dev训练模型模型和仓库中提供的性能差距甚远，请问作者是否了解这种情况或者是否解决过类似issue？

uuuzk

Optimize Laplacian computation for large matrices (significant speedup)

1

## Description The current Laplacian computation in `get_laplacian` is very slow for large matrices. For example, with `M` of size **16,000 × 16,000**, the original implementation takes **~300 seconds**. I...

simon-cas

结果中的“lm"是什么

1

感谢开源！请问图中的lm后缀的模型是什么意思？

dongska

jinhonglu