Yunchao He

Low GPU usage

17

I train the model on my two Tesla M40 GPU. When I use _nvidia-smi_ command to check the GPU usage, it always keeps on a low usage, and just one...

Difference Between Current Code and Original Paper

1

1. `Learning rate decay`. In the original paper, the learning rate decay starts from 0.001 and is reduced to 0.0005, 0.0003, and 0.0001 after 500K, 1M and 2M global steps...

post-processing tech on the generated waves

12

LPCNet performs very well on the features from ground-truth waves. However, when the features are predicted using an end2end model or SPSS model, there are some audible artifacts in the...

According to the paper, FB-MelGAN (F3) has a small noticeable increase in parameter size, computation complexity and real-time factor. It's very exciting that MB-MelGAN (M2) can generate high quality speech...

Optimize GPT2 inference: Remove redundant `autoregressive_latent_graph` and enable streaming output

3

Thank you for this excellent implementation. I'd like to suggest an optimization that could significantly speed up inference and enable streaming output. Currently, there are two GPT2 graphs: 1. autoregressive:...