[Proposal] Reduce training time by resampling beforehand

rishikksh20 / VocGAN

VocGAN: A High-Fidelity Real-time Vocoder with a Hierarchically-nested Adversarial Network

MIT License

319 stars 61 forks source link

[Proposal] Reduce training time by resampling beforehand #15

Open Craq opened 3 years ago

Craq commented 3 years ago

The biggest bottleneck in your model, regarding speed, is the Resample layer in the Hierarchical Discriminator. The speed on it is just awful. If you resample your dataset beforehand and load it from disk, your training will be up to 3 times faster.

rishikksh20 commented 3 years ago

I tried it but still not able to get good speed on baseline model.