Tatiana Likhomanenko

Please run with`--logtostderr=1` (to have printing all intermediate loggin on the screen) and set `--iters=100000` (we reported this several times in the issues on tutorial regarding our changes in the...

Hmm, strange, should work `--logtostderr=1` either you add into flagsfile or specify in cmd line.

cc @vineelpratap

Lexicon decoder for CTC is prefix beam search decoder. In the branch you pointed we try to continue with a token which different from the previous last token in the...

Or try to train/decode sota models from https://github.com/facebookresearch/wav2letter/tree/master/recipes/sota/2019.

For smearing we are using only unigrams scores. Unk word will use just unk score from kenlm (so kenlm has the unk word as normal word accumulating the full rest...

@trangtv57 yep, this is correct comment, kenlm has an option to restrict vocab. This restriction works as 2. pointed by @abhinavkulkarni: so this is done on fly during computation that...

cc @vineelpratap

Hey! In your log you can see that `I0911 11:03:37.456050 5675 Utils.cpp:102] Filtered 1/1 samples` which means that all samples are filtered. Could you run with `--decoder_nthreads=1 --maxisz=1000000` so that...

Could you try with `--maxisz=1000000 --maxtsz=1000000 --minisz=0 --minisz=0`?