-
Notifications
You must be signed in to change notification settings - Fork 0
Extending the HTK Speech Recognition toolkit from University of Cambridge for neural nets dropout training
License
lchanmann/HTK
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
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 0
No packages published