分享
012-Effect of nperspeaker/batchsize for BCE
输入“/”快速插入
012-Effect of nperspeaker/batchsize for BCE
飞书用户500
飞书用户2707
2023年7月12日创建
控制 batch每个人的样本数 batchBCE vs globalBCE
配置
batch_BCE_lam0.5_sampler
global_BCE_lam0.9_sampler
batch_size = 128
loss.sum
Clean test
Why increasing n_per_speaker, the performance reduced significantly?
1.
If bug? Check it (dataloader or loss modules)
2.
The convergence of loss?
The code of Batch_BCE_4 has some problems. Be solved and changed in loss.mean test.
Noisy test
For the Clean test, it can be observed that Batch_BCE outperforms Global_BCE. While for the Noisy test, Global_BCE wins Batch_BCE.
WHY?
loss.mean
Clean test