Skip to content

Latest commit

 

History

History
172 lines (122 loc) · 5.38 KB

INSTALL.rst

File metadata and controls

172 lines (122 loc) · 5.38 KB
Raw

Installation of Alex and the dependencies

This document describes how to install all dependencies needed to use Alex. The Alex project is developed in Python and tested with version 2.7. It may be necessary to have exactly this version of Python for the project to work correctly.

Ubuntu 14.04

Ask the root on the computer to run:

sudo apt-get install gfortran libatlas-base-dev portaudio19-dev swig \
    flac speex sox libsox-dev mplayer libsqlite3-dev python-wxgtk2.8 libmad0-dev \
    libjpeg8-dev libfreetype6-dev libpng12-dev libagg-dev libatlas3-base \
    libsox-fmt-mp3

To get latest versions of the following python packages, I recommend to run these commands:

sudo pip install --upgrade -r alex-requirements.txt
# The following step is optional as it installs dependencies specific for performing
# only certain tasks (like evaluating VAD performance).
sudo pip install --upgrade -r alex-requirements-dev.txt
sudo easy_install pysox

See alex-dsg/alex-requirements.txt.

Source code compiled packages

pyAudio

Get a special version of pyAudio from https://github.com/bastibe/PyAudio (bastibe-PyAudio-2a08fa7). This version supports non-blocking audio.

git clone https://github.com/bastibe/PyAudio.git
cd PyAudio
sudo python ./setup.py install

flite

Get the latest flite from http://www.festvox.org/flite/download.html and build it by following the these commands:

wget http://www.festvox.org/flite/packed/flite-1.4/flite-1.4-release.tar.bz2
tar -xvjf flite-1.4-release.tar.bz2
cd flite-1.4-release
./configure
make

Copy the flite-1.4-release/bin/flite file into you search path. E.g. link the flite program to your bin directory.

HTK

Get the latest HTK (3.4.1 tested) from http://htk.eng.cam.ac.uk/download.shtml . Build and install the HTK following the HTK's instructions.

KALDI

We are using our custom code based on Kaldi stored in separate project https://github.com/UFAL-DSG/pykaldi. The repository itself contains kaldi and pyfst (https://github.com/UFAL-DSG/pyfst) as submodules. Pykaldi implements also standalone C++ recogniser but the main focus is to prepare Python wrapper of the recogniser for Alex. The python wrapper is supported for Ubuntu 14.04 (the C++ decoder is as platform independent as Kaldi). For debugging installation see pykaldi/Makefile and its install target. Note: If installations instructions are not working for you try to look at pykaldi/Docker file, and let us know.

git clone https://github.com/UFAL-DSG/pykaldi

# install system prerequisities
sudo apt-get update && sudo apt-get install -y build-essential libatlas3-base libatlas-base-dev python-dev python-pip git wget gfortran
# install python prerequisities
sudo pip install -r pykaldi/pykaldi-requirements.txt

cd pykaldi
sudo make install
# for local installation
make deploy  # and add generated libraries to system paths and install generated eggs.

SRILM

Get the latest SRILM (1.6 tested) from http://www.speech.sri.com/projects/srilm/ . Build and install the SRILM following their instructions.

pjsip

Get the supported pjsip 2.1 from our fork at GitHub. To install pjsip, follow these instructions:

git clone git@github.com:UFAL-DSG/pjsip.git
cd pjsip
./configure CXXFLAGS=-fPIC CFLAGS=-fPIC LDFLAGS=-fPIC CPPFLAGS=-fPIC
make dep
make
sudo make install

then

cd pjsip-apps/src/python/
sudo python setup-pjsuaxt.py install

This will install the pjsuaxt library.

Morfodita

Get the supported morfodita from GitHub. To install morfodita, follow these instructions:

git clone git@github.com:ufal/morphodita.git
cd morphodita/src
make
cd ../bindings/python
make PYTHON_INCLUDE=/usr/include/python2.7/

sudo cp -R ./ufal /usr/local/lib/python2.7/dist-packages
sudo cp ./ufal_morphodita.so /usr/local/lib/python2.7/dist-packages

Optimised ATLAS and LAPACK libraries

If you need optimised ATLAS and LAPACK libraries then you have to compile them on your own. Then modify config for numpy. Optimised ATLAS and LAPACK can compute matrix multiplication on all CPU cores available.

To build your own optimised ATLAS and LAPACK libraries:

  • get latest LAPACK
  • get latest ATLAS
  • compile lapack
  • tell atlas where is your compiled LAPACK
  • compile ATLAS

Local installation of Python 2.7 and its dependencies

If you do not have the root access to the machine then you then you can use https://github.com/akheron/multipy to install the 2.7 version of Python and consequently to install all Python dependencies locally.

You can use the following script

multipy install 2.7

to download, compile, and install python 2.7 into ~/multipy directory.

To enable this local version, you have to call from your shell command line

source ~/multipy/pythons/2.7/bin/activate

You can also add the previous line into .bashrc to activate your local version of Python every time you start a bash console.

When you activate your local Python, you can install all python packages using pip or easy_install locally.