what brings the improvement of inference speed from 1200khz to 2750khz?

[wenyong-h]

The recent commit f70678e claims a big improvement of inference speed.
But there was no code change, so i am wondering what brings the improvement.

[yoyololicon]

Maybe the support of FP16 ?

[rafaelvalle]

We trained a new model that reduces the number of channels from 512 to 256 and still maintains the audio quality.

[hdmjdp]

@rafaelvalle have your try to reduce the WN layers(n_layers) from 8 to 4?

[rafaelvalle]

No, we have not.

[TomGledhill]

Hello @rafaelvalle! Thank you all, for this awesome project!
Could you please share pretrained weights of model with reduced number of channels?
Thanks in advance!

[tugstugi]

@rafaelvalle I second @TomGledhill is it possible to share the model with 256 channels?

[rafaelvalle]

Here is the link for the model with 256 channels trained on LJS.
https://drive.google.com/open?id=1WsibBTsuRg_SF2Z6L6NFRTT-NjEy1oTx