Ryan Hileman
Ryan Hileman
I just implemented Raymond Chen's [How can I launch an unelevated process from my elevated process and vice versa](https://devblogs.microsoft.com/oldnewthing/20131118-00/?p=2643) using windows-rs, which required VARIANT for the ShellExecute call. It took...
https://github.com/haya14busa/goverage/blob/master/main.go#L159-L162
No need to talk down to me. Bad RSI. I'll provide a PR tonight if this isn't enough. It can be effectively a single println in the err case. Harder...
> This repo uses MIT license so we can't just bundle something that is Apache 2.0 licensed You can certainly ship a wheel that has both MIT and Apache 2.0...
My [wav2train](https://github.com/facebookresearch/wav2letter/issues/509) project can align arbitrary audio for use with wav2letter
The wvalid command in [wav2train](https://github.com/talonvoice/wav2train) filters bad audio clips from a lst file.
The TDS CTC model is different from streaming convnets: https://github.com/facebookresearch/wav2letter/blob/master/recipes/models/sota/2019/am_arch/am_tds_ctc.arch https://github.com/facebookresearch/wav2letter/blob/master/recipes/models/sota/2019/librispeech/train_am_tds_ctc.cfg TDS CTC is 1.6GB, while sconv is only 400MB. Your results look fine for the data size you put...
What's your batch size?
I think zamia speech is an OG 1.6gb conv_glu asg model
FWIW I've found wav2letter models perform better with short inputs if you train them on short inputs. My high-performing english models have been trained on hundreds of thousands of clips...