deepspeech 0.0.2

Rust bindings to Mozilla's DeepSpeech STT implementation
Documentation

deepspeech-rs

crates.io

docs.rs

Rust bindings of Mozilla's DeepSpeech library.

The library is open source and performs Speech-To-Text completely offline. They provide pretrained models for English.

Quickstart

Preparation:

  1. git clone https://github.com/mozilla/DeepSpeech
  2. cd into that directory
  3. git checkout v0.1.0
  4. Run util/taskcluster.py to download precompiled components. See the native_client README for further options for that script and how to compile DeepSpeech yourself.
  5. Download the pretrained models from the URL https://github.com/mozilla/DeepSpeech/releases/download/v0.1.0/deepspeech-0.1.0-models.tar.gz and extract the zip file to some location.
  6. Add the directory where the precompiled components lie (the DeepSpeech checkout) to your LD_LIBRARY_PATH and LIBRARY_PATH environment variables.

You can now invoke the example via:

cargo run --release --example client <path-to-model-dir> <path-to-audio-file>

It will print out the recognized sequence on stdout. The format of the audio files is important: only mono files are supported for now. The 0.1.0 release announcement has a detailed list of requirements. All codecs that the awesome audrey library supports are supported.

Note: Right now, there are no Linux x64 binaries for older CPU platforms without AVX2 support like Ivy Bridge. See DeepSpeech's 0.1.0 release announcement for a list of supported platforms.

Supported versions of DeepSpeech

As of writing this, there has been only been one release of the DeepSpeech library yet, version 0.1.0. We only claim compatibility with that release. We will try to provide compatibility with the most recent usable release possible.

License

Licensed under Apache 2 or MIT (at your option). For details, see the LICENSE file.

All examples inside the examples/ folder are licensed under the CC-0 license.

License of your contributions

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed / CC-0 licensed as above, without any additional terms or conditions.