Skip to content
Snippets Groups Projects
Commit 68c5c058 authored by Daniel Povey's avatar Daniel Povey Committed by Dan Povey
Browse files

Some minor changes and renaming of files; starting documentation

parent c4d67a74
No related branches found
No related tags found
No related merge requests found
File moved
// lat/ctc-supervision.h
// lat/cctc-supervision.h
// Copyright 2015 Johns Hopkins University (Author: Daniel Povey)
......@@ -34,7 +34,9 @@
namespace kaldi {
// CTC means Connectionist Temporal Classification, see the paper by Graves et
// al.
// al. CCTC means context-dependent CTC, which
//
//
// What we are implementing here is some things relating to computation of the
// CTC objective function. The normal information required to compute the
......@@ -201,9 +203,9 @@ void AddBlanksToProtoSupervision(const CtcSupervisionOptions &options,
- Exactly one copy of that phone symbol.
- Respectively (before and after) each real phone symbol mentioned above,
optional repeats of the blank symbol (symbol 0). The time information for these
instances of the blank symbol will be the same as the phone symbol we
generated it from.
optional repeats of the blank symbol (symbol 0). The time information
for these instances of the blank symbol will be the same as the phone
symbol we generated it from.
This fst will have a state 0 which is both initial and final. Let's
suppose we have a phone-in-context a. first an arc from state 0 to a new
......
// doc/ctc.dox
// Copyright 2015 Johns Hopkins University (author: Daniel Povey)
// See ../../COPYING for clarification regarding multiple authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
// http://www.apache.org/licenses/LICENSE-2.0
// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
// MERCHANTABLITY OR NON-INFRINGEMENT.
// See the Apache 2 License for the specific language governing permissions and
// limitations under the License.
namespace kaldi {
/**
\page ctc Connectionist Temporal Classifiction
\section Introduction
This page describes the support for Connectionist Temporal Classifiction (CTC)
in Kaldi. At the current time CTC is only supported in the nnet3 setup, but
the non-neural-net parts of the CTC code are designed to be independent of the
neural net implementation so porting it to other setups such as nnet1 and nnet2
should be possible.
The original CTC paper is "Connectionist temporal classification: labelling
unsegmented sequence data with recurrent neural networks" by Alex Graves et al.
The core of CTC is somewhat indepenent of of the exact network topology
Originally the idea of CTC was to have the network output words directly, but
more recent work by Google, e.g.
"Fast and Accurate Recurrent Neural Network Acoustic Models for Speech
Recognition" by Hasim Sak, Andrew Senior et al.
Todo: finish this intro.
*/
}
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment