Struggling with installing/running Deepspeech

Archives DeepSpeech
1

Hi,

although training and testing with pre trained models works using gpu,
I cannot make it run for training and testing,
issues that pop up and and that I cannot fix, in spite of reading comments on similar issues:

issue 1
DeepSpeech/util/audio.py:17: DeepSpeechDeprecationWarning: DeepSpeech Python bindings could not be imported, resorting to slower code to compute audio features. Refer to README.md for instructions on how to install (or build) the DeepSpeech Python bindings.

issue2
_tensorflow.python.framework.errors_impl.NotFoundError: native_client/libctc_decoder_with_kenlm.so: undefined symbol: ZN10tensorflow20OpKernelConstruction21CtxFailureWithWarningEPKciRKNS_6StatusE

issue3
virtenv_deepspeech2/lib/python3.5/site-packages/matplotlib/init.py:1066: UserWarning: Duplicate key in file “/home/getalp/desotth/.config/matplotlib/matplotlibrc”, line #2
(fname, cnt))

my installation steps on a server with GPUs:

virtualenv -p python3 virtenv_deepspeech2
cd virtenv_deepspeech
source bin/activate
pip install deepspeech
pip install deepspeech_gpu
git clone https://github.com/mozilla/DeepSpeech.git
cd DeepSpeech
pip install -r requirements.txt
pip install tensorflow_gpu==1.4.0
(also tried : pip install tensorflow_gpu==1.5.0)
python util/taskcluster.py --arch gpu --target native_client

./bin/run-ldc93s1.sh

I would be really very grateful if someone can help me here, or is prepared to share more info.

Thanks!

2

This is clearly a mismatch betewen the tensorflow binaries and your libctc_decoder_with_kenlm.so.

3

This means you are working on master, so please respect requirements.txt and use TensorFlow v1.6.0.

Your util/taskcluster.py call downloads the latest (i.e., master) binaries. So it’s completely normal that you run into the symbol issue.

4

ok thanks !!!
runs now!!!

1 Like