Closed DatGuy1 closed 4 years ago
I'll update it soon (tm). I need to update the inference server a little as the text preprocessor keeps letting chunks larger than the target segment size through which is invalidating my new "Failure Rate" metric/score
example output from the inference server
Generated 2.566s of audio in 1.161s wall time - so far. (best of 64 tries this pass) (2.21xRT) (0.0%Failure Rate)
The failure rate is measured as a file with weighted attention score under 0.6. Every generated clip that doesn't make it to the end of it's input text easily gets written off as a failure clip, so that's currently being resolved.
GTA.py updated, will start updating HiFi-GAN dataloader when this is done processing.
Problem resolved. Closing issue.
Hey, unless I'm missing something you never actually updated GTA.py? Never mind. I found you updated it in tacotron2_tm rather than _3_generate_postnets.
ah yeah sorry, I'm slowly phasing our the _1_preprocess
and _3_generate_postnets
folders.
Still seems to use the training format that's in master.