Skip to content
Please note that GitHub no longer supports your web browser.

We recommend upgrading to the latest Google Chrome or Firefox.

Learn more

/wav2letter

Facebook AI Research's Automatic Speech Recognition Toolkit
wav2letter speech-recognition end-to-end deep-learning cpp
  1. C++ 81.1%
  2. CMake 9.1%
  3. Python 5.3%
  4. Cuda 4.2%
  5. C 0.3%
Branch: master
Find File
Clone or download

Clone with HTTPS

Use Git or checkout with SVN using the web URL.

Download ZIP

Launching GitHub Desktop...

If nothing happens, download GitHub Desktop and try again.

Launching GitHub Desktop...

If nothing happens, download GitHub Desktop and try again.

Launching Xcode...

If nothing happens, download Xcode and try again.

Launching Visual Studio...

If nothing happens, download the GitHub extension for Visual Studio and try again.

@jacobkahn @facebook-github-bot
jacobkahn and facebook-github-bot Make building w2l with FB dependencies configurable with Buck 
Summary:
Create an extra Buck config: `w2.build_fb_dependencies`, a bool flag for building wav2letter with FB dependencies (e.g. Everstore reader and all related deps). We've established several reasons for wanting this:
- Reduce FB build times when rebasing or building from scratch
- Don't rely on folly/other components which sometimes break in fbcode/break our build as a result
- Still be able to build in gcc if we want even though a variety of other internal FB projects are no longer explicitly supporting it

By default, FB deps aren't included:
```
buck build mode/opt deeplearning/projects/wav2letter:train_cpp
```
is much faster after pulling from master. To include Everstore reader/FB deps, use:
```
buck build mode/opt deeplearning/projects/wav2letter:train_cpp -c w2l.build_fb_dependencies=true
# different truthy/falsy values are respected, e.g. true, True, false, False, 0, 1, etc
```

Also renames the preprocessor macro to `W2L_BUILD_FB_DEPENDENCIES` to be more explicit

Since `read_bool` is the only configuration option for Buck that reads and casts bools reasonably (and it's only usable in Bazel files), create a separate config for the function.

Reviewed By: vineelpratap

Differential Revision: D15732262

fbshipit-source-id: 9a6b4214deb24a6a545400e8902bfd47644c246c
Latest commit cf0ee9f Jun 10, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.circleci CircleCI automated build checks and CUDA/CPU Dockerfile base May 31, 2019
cmake Automatically get/build gtest/gmock if building tests Jan 30, 2019
docs Update decoder doc Jun 4, 2019
recipes Update decoder doc Jun 4, 2019
src Make building w2l with FB dependencies configurable with Buck Jun 10, 2019
tutorials Update decoder doc Jun 4, 2019
.clang-format Bring back clang-format Apr 19, 2019
.gitignore Minor nits May 21, 2019
CMakeLists.txt Build stubs for experimental, semisupervised, ConvLM Jun 10, 2019
CODE_OF_CONDUCT.md Initial Commit Dec 21, 2018
CONTRIBUTING.md Bring back clang-format Apr 19, 2019
Decode.cpp Fix trivial bugs in Test.cpp and Decode.cpp Jun 7, 2019
Dockerfile-CPU CircleCI automated build checks and CUDA/CPU Dockerfile base May 31, 2019
Dockerfile-CPU-Base CircleCI automated build checks and CUDA/CPU Dockerfile base May 31, 2019
Dockerfile-CUDA CircleCI automated build checks and CUDA/CPU Dockerfile base May 31, 2019
Dockerfile-CUDA-Base CircleCI automated build checks and CUDA/CPU Dockerfile base May 31, 2019
LICENSE Initial Commit Dec 21, 2018
README.md CircleCI automated build checks and CUDA/CPU Dockerfile base May 31, 2019
Test.cpp Fix trivial bugs in Test.cpp and Decode.cpp Jun 7, 2019
Train.cpp rename some helper functions Jun 6, 2019

README.md

wav2letter++

CircleCI

wav2letter++ is a fast open source speech processing toolkit from the Speech Team at Facebook AI Research. It is written entirely in C++ and uses the ArrayFire tensor library and the flashlight machine learning library for maximum efficiency. Our approach is detailed in this arXiv paper.

The goal of this software is to facilitate research in end-to-end models for speech recognition.

The previous version of wav2letter (written in Lua) can be found in the "wav2letter-lua" branch under the repository.

Building wav2letter++

See Building Instructions for details.

Full documentation

To get started with wav2letter++, checkout the tutorials section.

We also provide complete recipes for WSJ, Timit and Librispeech and they can be found in recipes folder.

Citation

If you use the code in your paper, then please cite it as:

@article{pratap2018w2l,
  author          = {Vineel Pratap, Awni Hannun, Qiantong Xu, Jeff Cai, Jacob Kahn, Gabriel Synnaeve, Vitaliy Liptchinsky, Ronan Collobert},
  title           = {wav2letter++: The Fastest Open-source Speech Recognition System},
  journal         = {CoRR},
  volume          = {abs/1812.07625},
  year            = {2018},
  url             = {https://arxiv.org/abs/1812.07625},
}

Join the wav2letter community

See the CONTRIBUTING file for how to help out.

License

wav2letter++ is BSD-licensed, as found in the LICENSE file.

You can’t perform that action at this time.