Kaldi Speech Recognition Toolkit – Designed for Speech Recognition Researchers

Kaldi is a state-of-the-art speech recognition toolkit written in C++. It’s intended to be used mainly for acoustic modelling research.

Kaldi provides a speech recognition system based on finite-state transducers (using the freely available OpenFst), together with detailed documentation and scripts for building complete recognition systems. Its core library supports modelling of arbitrary phonetic-context sizes, acoustic modelling with subspace Gaussian mixture models (SGMM) as well as standard Gaussian mixture models, together with all commonly used linear and affine transforms.

Even though the software has been in development for some years, it’s in an early stage of development; there are no official releases of the software.

Features include:

  • Code-level integration with Finite State Transducers (FSTs) – compiles against the OpenFst toolkit.
  • Tools for converting LMs in the standard ARPA format to FSTs.
  • Extensive linear algebra support including a matrix library that wraps standard Basic Linear Algebra Subroutines (BLAS) and Linear Algebra Package (LAPACK) routines.
  • An extensible design.
  • MMI.
  • Boosted MMU.
  • MCE discriminative training.
  • Feature-space discriminative training (similar to fMPE, but based on boosted MMI).
  • Deep neural networks.
  • Supports Gaussian mixture modules with diagonal and full covariance structures.
  • Support both model-space adaptation using maximum likelihood linear regression (MLLR) and feature-space adaptation using feature-space MLLR (fMLLR), also known as constrained MLL.
  • Complete recipes.

OpenFst is a library for constructing, combining, optimizing, and searching weighted finite-state transducers (FSTs). Weighted finite-state transducers are automata where each transition has an input label, an output label, and a weight. The more familiar finite-state acceptor is represented as a transducer with each transition’s input and output label equal. Finite-state acceptors are used to represent sets of strings (specifically, regular or rational sets); finite-state transducers are used to represent binary relations between pairs of strings (specifically, rational transductions). The weights can be used to represent the cost of taking a particular transition.

Website: kaldi-asr.org
Support: GitHub Code Repository
Developer: Daniel Povey, Arnab Ghoshal, Gilles Boulianne, and many others
License: Apache License v2.0

Kaldi is written in C++. Learn C++ with our recommended free books and free tutorials.

Return to Speech Recognition Tools Home Page

Read our complete collection of recommended free and open source software. The collection covers all categories of software.

The software collection forms part of our series of informative articles for Linux enthusiasts. There's tons of in-depth reviews, alternatives to Google, fun things to try, hardware, free programming books and tutorials, and much more.
Share this article

Share your Thoughts

This site uses Akismet to reduce spam. Learn how your comment data is processed.