Presentation

ALIZÉ is an opensource platform for speaker recognition. The purpose of this project is to provide a set of low-level and high-level frameworks that will allow anybody to develop applications handling the various tasks in the field of speaker recognition: verification/identification, segmenting, etc.

Structure of the platform

In order to better suit everybody’s needs, ALIZÉ was developed with a multi-layer architecture.

The base layer is ALIZE-Core, a low-level library which includes all the functions required to use Gaussian mixtures, as well as I/O functions for various file formats.

On top of this core was built LIA_RAL, a toolkit offering higher-level functionnality. LIA_RAL is itself made of several components:

LIA_SpkDet
A set of tools to do all the tasks required by a speaker authentication system — model training, feature normalization, score normalization, etc.
LIA_SpkSeg
Tools for speaker diarization.
LIA_Utils
Utilities for manipulating the various data formats used in ALIZÉ — GMMs, features, etc.
LIA_SpkTools
The library on which the other parts are based; it provides high-level functions on top of ALIZE-core.

In parallel, LIA_RAL also includes a library named SimpleSpkDetSystem, which offers a simple, high-level API for developers who want to easily embed speaker verification or identification in their applications. A Java version of this API was also developed, targeting Android application development.

Compatibility

This software was developed with multi-platform compatibility in mind. It has been tested and used on Linux, Windows, and Mac OS. It should also be usable on any POSIX-compliant operating system.

Through the Android-ALIZÉ project, ALIZÉ can also be embedded in mobile applications running on the Android platform.

Download ALIZÉ

The source code of ALIZÉ is available on GitHub at this address: https://github.com/ALIZE-Speaker-Recognition.

License

The source code of ALIZÉ (both ALIZE-Core and LIA_RAL) is released under the terms of the GNU Lesser General Public License (LGPL).

Complementary Software

ALIZÉ does not include functions for feature extraction from the speech signal. For this task, it relies on the free speech signal processing toolkit SPro, developed by Guillaume Gravier at IRISA: https://gforge.inria.fr/projects/spro/.

⚠️ Note that only the revisions 155 and up of SPro are fully compatible with 64-bit CPUs. However, at the time of this writing, these versions of SPro are only available through Subversion, and the direct download link given on the website above points to an older revision of SPro 5 which includes a bug leading to corrupted feature files when compiled for 64 bit systems.

If you want to be sure to get the right version of SPro for use with ALIZÉ in 64-bit mode, you can download it from here: spro-5.0-157.tar.gz.

How to use ALIZÉ

API Documentation

Unfortunately, the documentation for the library APIs is in dire need of an update, and is therefore not available for the current version of ALIZÉ.

Tutorials

Several tutorials are available to get you started with the tools LIA_SpkDet and LIA_SpkSeg. For convenience, the archives below include pre-compiled binaries for Linux on x86-64. If you are on a different platorm, or if you want to look at the source code, please go to the download section in order to get the source code for ALIZÉ.

Community

Since the launch of the project in 2005, ALIZÉ has attracted a rich community of users and developers.

The best way to exchange with the community is through the dev-alize mailing list:

The community may also be reached through the ALIZÉ group on LinkedIn.

Publications

Follow this link to see a list of scientific papers related to ALIZÉ and its use for research in speaker recognition.

Contribute

ALIZÉ is opensource software. If you want to extend it or submit bug fixes, do not hesitate to join the community and help us maintain a state-of-the-art speaker recognition toolkit.

Of course, there are also other ways you can contribute:

Feel free to join the effort! :-)