-
We need a new training script, similar to steps/nnet3/chain/train.py for training with data from multiple languages. The changes required are as follows:
1. Generate chain egs for multiple languages (…
-
It just occurred to me that when training the left-biphone chain model, it would be interesting to do a pass of SAT training with a left-biphone model, including alignment passes. This may improve th…
-
Change the tdnn config-generation script so that if you specify pnorm input or output dim, it forces p-norm nonlinearity, or at least dies with an error if you're using relu.
-
User vince62s has updated the TED-LIUM recipe to use the TED-LIUM 2 release of data, which almost doubles the data. Obviously we want to add these changes to the recipe. However we would like to prese…
-
One of the footnotes of the paper states that the character-level representation matrix C-k is zero-padded so that the input to TDNN all have fixed size like batch_size x max_word_l x char_vec_size. I…
-
Using the latest data-augmentation scripts which use the `wav-reverberate` command I am unable to replicate the results in the ASpIRE recipe. When I use the old data-augmentation scripts from around M…
-
In the current chain recipes we use a tri-phone LM. We would like to test out left-biphone LM in its place. @keli78 has tested these LMs in RM recipe and found them helpful. We would like to test it …
-
Trying to run main.lua, I'm getting the following error:
```
Word vocab size: 599589, Char vocab size: 42, Max word length (incl. padding): 65
creating an LSTM-CNN with 2 layers
/home/robin/dev/torc…
-
I get this, running nnet3 decoding with --iter param
steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 --cmd queue.pl --iter 200 --nj 16 exp/chain/tdnn_2o_4/graph/ data/test_hires/ exp/chain/t…
-
Hi:
The previous version using Squeeze.lua works fine.
I would like to test sample.lua but it does not work with the previous version.
I'm new to lua/torch and I would like to sample from this mod…
ghost updated
8 years ago