Skip to content
/ HTK Public

Extending the HTK Speech Recognition toolkit from University of Cambridge for neural nets dropout training

License

Notifications You must be signed in to change notification settings

lchanmann/HTK

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

!!! IMPORTANT !!!
=================
Please register at https://htk.eng.cam.ac.uk/register.shtml before you
continue reading/using the code here.

The Hidden Markov Toolkit (HTK) is made available free of charge and can
be downloaded from the HTK website. However it may not be redistributed,
i.e. you must register at the website and download it from there. 


Hidden Markov Toolkit (HTK) 3.5 Beta
====================================

Use of this software is governed by a license agreement, the terms and
conditions of which are set forth in the file LICENSE in the
top-level HTK installation directory.  Please read this file carefully
as use of this software implies acceptance of the conditions described
therein.


Introduction
============

HTK is a toolkit that has been primarily for research use in automatic 
speech recognition.  Since 2000, HTK has been developed and distributed 
by the Machine Intelligence Laboratory (formerly known as the Speech Vision 
and Robotics Group) at Cambridge University Engineering Department 
(http://mi.eng.cam.ac.uk).

Please visit the HTK website at the following address for more
information about HTK and the history of its development:

    http://htk.eng.cam.ac.uk/

A number of mailing lists have been established to help users build
and understand HTK, for details see

    http://htk.eng.cam.ac.uk/mailing/subscribe_mail.shtml


License
=======

HTK is made available free of charge and can be downloaded from the
HTK website. However it may not be redistributed, i.e. you must 
register at the website and download it from there. 

Details about the terms under which HTK is made available can
be found in the LICENSE file.

Note that the HDecode tool must be downloaded separately and
has a separate license agreement.


Compiling & Installing HTK under UNIX/Linux, OS X or Cygwin
===========================================================

HTK 3.5 comes with three sets of Makefiles enabling compilation for:

1. CPU (standard)
2. CPU using MKL (Intel Math Kernel library)
3. NVIDIA GPUs (Graphic Processing Unit) using the NVIDIA CUDA compiler NVCC

These alternative Makefiles are located in each of the following directories:

HTKLib, HLMLib, HTKTools, HLMTools

Examine the MakefileCPU, MakefileMKL and  MakefileNVCC 
files in each of these directories and change the default 
locations of CPU/GPU compilers and libraries to match the
locations on the system for which HTK 3.5 is to be installed.

In order to compile either the standard CPU version, or 
the CPU version with MKL support or GPU version execute one of the following

1. make -f MakefileCPU all
2. make -f MakefileMKL all 
3. make -f MakefileNVCC all

in turn in each of HTKLib, HLMLib, HTKTools and HLMTools.
Having compiled one of the branches (CPU/MKL/NVCC), the 
relevant HTK tools can be installed executing one of the following

1. make -f MakefileCPU install
2. make -f MakefileMKL install
3. make -f MakefileNVCC install

commands in both the HTKTools and HLMTools directories. By default HTK 
tools will be installed in bin.cpu, bin.mkl or bin.gpu depending 
on the branch. 

Prior to compiling a new branch it is important to clean the distribution

1. make -f MakefileCPU clean
2. make -f MakefileMKL clean
3. make -f MakefileNVCC clean


HDecode
=======

If you are also building HDecode (available from the HTK website, under a
different licence from HTK), you will firstly need to unpack the HDecode
source code (in the same directory in which you unpacked the HTK
sources). Then follow the steps above for first building HTK, and then execute 
the following command in HTKLVRec directory:

1. make -f MakefileCPU all
2. make -f MakefileCPU install

or

1. make -f MakefileMKL all
2. make -f MakefileMKL install

or

1. make -f MakefileNVCC all
2. make -f MakefileNVCC install

depending on whether HTK is compiled for CPU, CPU with MKL support or GPU.


Compiling & Installing HTK under Windows
========================================
HTK 3.5 has not yet been tested under Windows. 


Testing the Installation
========================
As an initial test of the installation please run the HTK demonstration 
using the configuration file samples/HTKDemo/configs/monPlainM1S1.dcf. 
There is a README file in the samples/HTKDemo directory that explains 
the operation of the demonstration in detail but, in short, you need 
to run the demonstration script passing it the configuration file 
samples/HTKDemo/configs/monPlainM1S1.dcf as input. 
To test the language modelling tools you should follow the tutorial
in the HTK book, using the files in the LMTutorial/ directory.

In addition to basic testing of the HTK installation, it is also possible 
to build several HTK systems for the Resource Management (RM) task. 
These systems range in complexity from simpler Gaussian mixture model 
based HMMs to more complex artificial neural network based systems. 
The RM recipe is located in samples/RMHTK directory and is documented 
as a part of tutorial chapter in the HTK book.

About

Extending the HTK Speech Recognition toolkit from University of Cambridge for neural nets dropout training

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published