diff options
| author | Thomas Mesnard <thomas.mesnard@ens.fr> | 2015-12-28 20:51:50 +0100 |
|---|---|---|
| committer | Alex Auvolat <alex@adnab.me> | 2016-04-21 10:21:42 +0200 |
| commit | 072d26e766931007a0f243674f7dfdff5c3104e9 (patch) | |
| tree | ae3639f4ff3f8e0e3e9767c15322171aa6f2169e /README.md | |
| parent | e8e37dee0c5c846b1aa2dd24dc99095191f72a9b (diff) | |
| download | pgm-ctc-072d26e766931007a0f243674f7dfdff5c3104e9.tar.gz pgm-ctc-072d26e766931007a0f243674f7dfdff5c3104e9.zip | |
Add plot
More TIMIT ; log domain
TIMIT: more complexity
Nice poster
Beautify code (mostly, add comments)
Add final stuff.
Diffstat (limited to 'README.md')
| -rw-r--r-- | README.md | 11 |
1 files changed, 9 insertions, 2 deletions
@@ -1,4 +1,11 @@ -# pgm -Projet PGM +# CTC implementation for Blocks and Theano Thomas Mesnard, Alex Auvolat + +This repository contains an implementation of the CTC cost function (Graves et al., 2006). To avoid numerical underflow, two solutions are implemented: + +- Normalization of the alphas at each timestep +- Calculations in the logarithmic domain + +This repository also contains sample code for applying CTC to two datasets, a simple dummy dataset constituted of artificial data, and code to use the TIMIT dataset. The model on the TIMIT dataset is able to learn up to 50% phoneme accuracy using no handcrafted processing of the signal, but instead uses an end-to-end model composed of convolutions, LSTMs, and the CTC cost function. + |