It is very slow for on the fly feature extract, how to solve it?
Naminwang opened this issue · comments
It is very slow for on the fly feature extract, how to solve it?
You could carry your point in a roundabout way. At first, using spafe,feature.lpc extract lpc feature, and then store them in torch.save(). This is still a CPU and off-line way, but it is correct absolutely.
For this project, Cos there are a lot details realized by CPU, so more optimizations is need. Hope I could do it in next month😂