Skip to content

DeepSpeech 0.6.0 for Jetson Nano

Compare
Choose a tag to compare
@domcross domcross released this 05 Dec 19:38
· 2 commits to master since this release
354c648

This DeepSpeech release requires Tensorflow 1.14. Luckily there is a official TF-release from Nvidia that works perfectly, so no compiling this time...

  1. Install Tensorflow 1.14 for Jetson Nano. Follow all instructions including prerequisites/requirements from there

  2. Download the DeepSpeech-0.6.0 wheel from this release, then pip install deepspeech-0.6.0-cp36-cp36m-linux_aarch64.whl

  3. Download the libdeepspeech.so file as well and put it in your library search path. (The actual location depends on your installation, e.g. python-venv)

When you want to build the wheel yourself:

  1. Follow the build instructions for the native-client (bazel, cloning Tf and DS, configure, ...)
  2. Apply patches for Tensorflow and DeepSpeech
  3. build the native client with following command (instead of the one given in the build instructions:
bazel build --config=opt --config=nonccl --config=noaws --jobs 4 --ram_utilization_factor 50 --verbose_failures --cxxopt="-D_GLIBCXX_USE_CXX11_ABI=0" --config=noignite --config=nokafka --copt=-O3 --define=target_system=rpi3-armv8 --copt=-march=armv8-a --copt=-mtune=cortex-a57 --copt=-DRASPBERRY_PI --config=nogcp --config=monolithic //native_client:libdeepspeech.so //native_client:generate_trie
  1. follow instruction for "Install Python bindings"
  2. done