Hello. I’m newbe.
I want to know exact word time using deepspeech. For example, when I use deepspeech, output is “I’m newbe …” just sentense. But, I want output like “I 1s am 2s newbe 4s…”. I knew the latest deepspeech had this function. But if I use this, do I have to create new .so file?
If you know that, it means you’ve read about the ctcdecode work, and you should know it’s not yet landed.
So if you want to play with it right now, yes, you have to build yourself.
Thank you for fast reply! Do you know how to make .so file? I found that when I download deepspeech using pip, there was just .so file. So, I tried to make .so file by downloading deepspeech open source and commend “sudo gcc -fPIC -c deepspeech.cc” but, there were many mis include file path, so I could not make it. And Can ctcdecode perform its technique correctly?
Please read the documentation in README.md and native_client/README.md.
Thank you for reply. So, I am following README.md.
I upgraded tensorflow, installed bazel and installed source code of branch “ctcdecode”.
My question is
bazel build -c opt --copt=-O3 --copt="-D_GLIBCXX_USE_CXX11_ABI=0" //native_client:libctc_decoder_with_kenlm.so
bazel build --config=monolithic -c opt --copt=-O3 --copt="-D_GLIBCXX_USE_CXX11_ABI=0" --copt=-fvisibility=hidden //native_client:libdeepspeech.so //native_client:generate_trie
I don’t know where to run that commend. I performed preparation step.
What do you mean, where? The Preparation instructs you to be in the TensorFlow source directory, and then create a symlink, you need to stay there to actually perform bazel build
My tensorflow path : /usr/local/lib/python3.6/dist-packages/tensorflow
My deepspeech path : /home/kwan/Downloads/DeepSpeech-ctcdecode/native_client
So preparation step :
cd tensorflow
ln -s /usr/local/lib/python3.6/dist-packages/tensorflow ./
And I conformed native_client in /usr/local/lib/python3.6/dist-packages/tensorflow.
And when I run the commend “sudo bazel build -c opt --copt=-O3 --copt=”-D_GLIBCXX_USE_CXX11_ABI=0" //native_client:libctc_decoder_with_kenlm.so",
result is
Extracting Bazel installation…
ERROR: The ‘build’ command is only supported from within a workspace.
WARNING: --batch mode is deprecated. Please instead explicitly shut down your Bazel server using the command “bazel shutdown”.
What is the problem?
Have you read the documentation ? You are using the Python tensorflow installation path, not the TensorFlow sources … And you are not applying the proper steps.
Yes, my fault…
By the way, last night I found that new deepspeech version was distributed.
I conformed that new version has ctcdecode. But, I upgrade my deepspeech yet.
Can you tell me the result? When I upgrade it, can I see the word and time?
We have not yet released anything, we just have the code to allow this feature. Please refer to https://github.com/mozilla/DeepSpeech/issues/1372