Eesen – End-to-End Speech Recognition using Deep RNN Models and WFST-based Decoding

The Eesen framework drastically simplifies the existing pipeline to build state-of-the-art ASR systems. Acoustic models in Eesen are deep bidirectional RNNs trained with the CTC objective function.

Eesen contains 4 key components to enable end-to-end ASR:

  • Acoustic Model — Bi-directional RNNs with LSTM units.
  • Training — Connectionist temporal classification (CTC) as the training objective.
  • WFST Decoding — A principled decoding approach based on Weighted Finite-State Transducers (WFSTs), or
  • RNN-LM Decoding — Decoding based on (character) RNN language models, when using Tensorflow.

Acoustic modeling in Eesen involves learning a single recurrent neural network (RNN) predicting context-independent targets (phonemes or characters). To remove the need for pre-generated frame labels, the connectionist temporal classification (CTC) objective function is adopted to infer the alignments between speech and label sequences. A distinctive feature of Eesen is a generalized decoding approach based on weighted finite-state transducers (WFSTs), which enables the efficient incorporation of lexicons and language models into CTC decoding. Experiments show that compared with the standard hybrid DNN systems, Eesen achieves comparable word error rates (WERs), while at the same time speeding up decoding significantly.

Features include:

  • The WFST-based decoding approach can incorporate lexicons and language models into CTC decoding in an effective and efficient way.
  • The RNN-LM decoding approach does not require a fixed lexicon.
  • GPU implementation of LSTM model training and CTC learning, now also using Tensorflow.
  • Multiple utterances are processed in parallel for training speed-up.
  • Fully-fledged example setups to demonstrate end-to-end system building, with both phonemes and characters as labels, following Kaldi recipes and conventions.

Website: github.com/srvk/eesen
Support:
Developer: Yajie Miao and contributors
License: Apache License 2.0

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

Return to Speech Recognition Tools Home Page


Ongoing series
Linux for StartersNew to Linux? Read our Linux for Starters series.
Free and Open Source SoftwareThe largest compilation of the best free and open source software in the universe. Supplied with our legendary ratings charts.
Linux ReviewsHundreds of in-depth reviews offering our unbiased and expert opinion on software.
Alternatives to GoogleAlternatives to Google's Products and Services examines your options to migrate from the Google ecosystem with open source Linux alternatives.
MicrosoftAlternatives to Microsoft's Products and Services recommends open source Linux software.
Linux System ToolsEssential Linux system tools looks at small, indispensable utilities, useful for system administrators as well as regular users.
Linux ProductivityLinux utilities to maximise your productivity. Small, indispensable tools, useful for anyone running a Linux machine.
Home Computer EmulatorsHome computers became commonplace in the 1980s. Emulate home computers including the Commodore 64, Amiga, Atari ST, ZX81, Amstrad CPC, and ZX Spectrum.
Now and ThenNow and Then examines how promising open source software fared over the years.
Linux at HomeLinux at Home looks at a range of home activities where Linux can play its part, making the most of our time at home, keeping active and engaged.
Linux CandyLinux Candy opens up to the lighter side of Linux. Have some fun!
Best Free Android AppsBest Free Android Apps. There's a strict eligibility criteria for inclusion in this series
Programming BooksThese best free books accelerate your learning of every programming language
Programming TutorialsThese free tutorials offer the perfect tonic to the free programming books series
Stars and StripesStars and Stripes is an occasional series looking at the impact of Linux in the USA
Share this article

Share your Thoughts

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