Batch size set to 1 - Githubissues

TzuchengChang / NASS

Noise-Aware Speech Separation with Contrastive Learning

14 stars 6 forks source link

Batch size set to 1 #4

Open hulucky1102 opened 3 months ago

hulucky1102 commented 3 months ago

Hi, why is the training batch size set to 1, and does this approach also work effectively for speech enhancement?

TzuchengChang commented 3 months ago

Due to the limitations of graphics memory at that time, we set the batch_size to 1. You can set it to other values. According to our tests, if you use NVIDIA V100 with 40GB graphics memory, the batch_size can be set to 4. This task can be used for speech enhancement.