Bug: Failed to resolve: ai.coqui:libstt:0.9.3 about stt

[reuben]

Hey @soudalan. That's right, currently we only have pre-built packages published for Python on Linux, macOS and Windows. We're working to bring the rest of the Continuous Integration system up to speed. Windows support has just been merged today, in fact. I'll close this bug once the Java builds are running and the package is available.

[soudalan]

Thanks for clarifying, looking forward to it! I assume in the meantime, using the old DeepSpeechDemo is my only alternative for Android then?

[reuben]

That, or building from source.

[soudalan]

How would I build from the source? Not sure if there was a wiki or page anywhere explaining the process.

[reuben]

Following the building docs until the configure step, then switch to the Android build steps which show how to build libstt.so and then libstt.aar. But as I'm taking a look now, I'm realizing the state of building from source is not ideal. You'll need to manually copy the built libstt.so into the correct libstt/libs folder, and then also edit native_client/java/libstt/gradle.properties to keep only the architecture you're interested in using, before you run the ./gradlew libstt:build step.

Hope this helps. If you try it and run into any problems please let me know so I can improve the docs. I'll also file an issue for improving the docs and the build system here to make it easier to build from source.

[soudalan]

I forgot to ask this earlier, but it seems I will need a GPU to build from the source? And assuming I don't have a GPU available, are there any other alternatives?

[reuben]

No, you don't need a GPU to build from source. What made you think that?

[soudalan]

Got it. The setup for Tensorflow asked for a path to my CUDA setup, so wasn't sure if I would also need a GPU prepared as well.

[reuben]

Ah, yes. You need CUDA installed to make GPU builds, but no actual GPUs are required. And you can also just not enable CUDA in the configure step to do CPU only builds.

[soudalan]

Under the "Building libstt.so for Android" section, when I run the command
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=android --config=android_arm --define=runtime=tflite --action_env ANDROID_NDK_API_LEVEL=21 --cxxopt=-std=c++14 --copt=-D_GLIBCXX_USE_C99 //native_client:libstt.so
, I get the error:

ERROR: Skipping '//native_client:libstt.so': no such package 'native_client': BUILD file not found in any of the following directories. Add a BUILD file to a directory to mark it as a package.

This is after the symlink command ln -s ../native_client.
What BUILD file do I need to add exactly?

[reuben]

If you created the symlink inside the tensorflow folder and if you're running bazel build from the tensorflow folder, that's all that should be needed.

[soudalan]

Hmm, that's what I just did but it's throwing the error still.

[reuben]

Are you on Windows by any chance?

[soudalan]

No, I'm on Mac OS (Big Sur)

[soudalan]

ERROR: Skipping '//native_client:libstt.so': no such package 'native_client': BUILD file not found in any of the following directories. Add a BUILD file to a directory to mark it as a package.

• /Users/jlam/Documents/coqui/tensorflow/native_client
  WARNING: Target pattern parsing failed.
  ERROR: no such package 'native_client': BUILD file not found in any of the following directories. Add a BUILD file to a directory to mark it as a package.
• /Users/jlam/Documents/coqui/tensorflow/native_client
  INFO: Elapsed time: 0.156s
  INFO: 0 processes.
  FAILED: Build did NOT complete successfully (0 packages loaded)
  jlam-macbookpro% ls
  ACKNOWLEDGMENTS LICENSE configure.py
  AUTHORS README.md models.BUILD
  BUILD RELEASE.md native_client
  CODEOWNERS SECURITY.md tensorflow
  CODE_OF_CONDUCT.md WORKSPACE third_party
  CONTRIBUTING.md arm_compiler.BUILD tools
  ISSUES.md configure
  ISSUE_TEMPLATE.md configure.cmd
  jlam-macbookpro% pwd
  /Users/jlam/Documents/coqui/tensorflow
  jlam-macbookpro%

[reuben]

Can you run, from the tensorflow folder, the following command and post their outputs?
ls -lha
ls -lha native_client

[soudalan]

jlam-macbookpro% ls -lha
total 1112
drwxr-xr-x  31 jlam  primarygroup   992B Apr 20 11:37 .
drwxr-xr-x   4 jlam  primarygroup   128B Apr 19 15:23 ..
-rw-r--r--   1 jlam  primarygroup    29K Apr 19 15:26 .bazelrc
-rw-r--r--   1 jlam  primarygroup     6B Apr 19 15:26 .bazelversion
drwxr-xr-x  13 jlam  primarygroup   416B Apr 19 15:27 .git
drwxr-xr-x   6 jlam  primarygroup   192B Apr 19 15:26 .github
-rw-r--r--   1 jlam  primarygroup   896B Apr 19 15:26 .gitignore
lrwxr-xr-x   1 jlam  primarygroup    34B Apr 19 15:26 .pylintrc -> tensorflow/tools/ci_build/pylintrc
-rw-r--r--   1 jlam  primarygroup   688B Apr 20 08:55 .tf_configure.bazelrc
-rw-r--r--   1 jlam  primarygroup   2.2K Apr 19 15:26 ACKNOWLEDGMENTS
-rw-r--r--   1 jlam  primarygroup   349B Apr 19 15:26 AUTHORS
-rw-r--r--   1 jlam  primarygroup    95B Apr 19 15:26 BUILD
-rw-r--r--   1 jlam  primarygroup   656B Apr 19 15:26 CODEOWNERS
-rw-r--r--   1 jlam  primarygroup   5.2K Apr 19 15:26 CODE_OF_CONDUCT.md
-rw-r--r--   1 jlam  primarygroup   9.7K Apr 19 15:26 CONTRIBUTING.md
-rw-r--r--   1 jlam  primarygroup   606B Apr 19 15:26 ISSUES.md
-rw-r--r--   1 jlam  primarygroup   2.2K Apr 19 15:26 ISSUE_TEMPLATE.md
-rw-r--r--   1 jlam  primarygroup    13K Apr 19 15:26 LICENSE
-rw-r--r--   1 jlam  primarygroup    21K Apr 19 15:26 README.md
-rw-r--r--   1 jlam  primarygroup   333K Apr 19 15:26 RELEASE.md
-rw-r--r--   1 jlam  primarygroup    13K Apr 19 15:26 SECURITY.md
-rw-r--r--   1 jlam  primarygroup   665B Apr 19 15:26 WORKSPACE
-rw-r--r--   1 jlam  primarygroup   1.1K Apr 19 15:26 arm_compiler.BUILD
-rwxr-xr-x   1 jlam  primarygroup   285B Apr 19 15:26 configure
-rw-r--r--   1 jlam  primarygroup   782B Apr 19 15:26 configure.cmd
-rw-r--r--   1 jlam  primarygroup    52K Apr 19 15:26 configure.py
-rw-r--r--   1 jlam  primarygroup   328B Apr 19 15:26 models.BUILD
lrwxr-xr-x   1 jlam  primarygroup    16B Apr 20 11:37 native_client -> ../native_client
drwxr-xr-x  36 jlam  primarygroup   1.1K Apr 20 11:35 tensorflow
drwxr-xr-x  97 jlam  primarygroup   3.0K Apr 19 15:27 third_party
drwxr-xr-x   4 jlam  primarygroup   128B Apr 19 15:29 tools
jlam-macbookpro% ls -lha native_client
lrwxr-xr-x  1 jlam  primarygroup    16B Apr 20 11:37 native_client -> ../native_client

[reuben]

Yeah it looks exactly like it should. And can you also do cat native_client/BUILD from the tensorflow folder? To be clear, that symlink should be pointing to the native_client folder inside the STT repo.

[soudalan]

It says

cat: native_client/BUILD: No such file or directory

[reuben]

Yeah, so either you're not in the right tensorflow folder (the submodule) or something in your system is broken in a way I've never seen. Make sure you have the following structure:

STT:
  - native_client/
    - BUILD
    - ...
  - tensorflow/
    - native_client -> ../native_client
    - ...
  - ...

[soudalan]

Good catch! Turns out I had two separate Tensorflow subdirecties, and I was running these commands inside the one that wasn't part of the STT directory. Thanks for pointing that out!

[soudalan]

I ran into this issue when building specifically for Android:

ERROR: /Users/jlam/Documents/STT/tensorflow/tensorflow/lite/kernels/internal/BUILD:687:1: C++ compilation of rule '//tensorflow/lite/kernels/internal:kernel_utils' failed (Exit 1)
src/main/tools/process-wrapper-legacy.cc:58: "execvp(/bin/false, ...)": No such file or directory
Target //native_client:libstt.so failed to build

This is after running

bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=android --config=android_arm --define=runtime=tflite --action_env ANDROID_NDK_API_LEVEL=21 --cxxopt=-std=c++14 --copt=-D_GLIBCXX_USE_C99 //native_client:libstt.so

I found this: tensorflow/tensorflow#2640

it looks like /bin/false is failing with no such file directory on execvp. That could be because Mac OS has /bin/false in /usr/bin/false instead. Could you try ln -s /usr/bin/false to /bin/false and see if it gets any further?

But that gave me an error:

ln -s /usr/bin/false /bin/false
ln: /bin/false: Operation not permitted

[reuben]

Did you try the second solution posted in that issue? tensorflow/tensorflow#2640 (comment)

[soudalan]

As this was posted 5 years ago, both versions are rather outdated now, so I wasn't sure if it made sense to build with such old versions. It could possibly still be an issue with version mis-matches, so I will try changing the version.

[NadeemAraman]

Please i need a firm answer , do i need to use Gpu (Cuda paths) to build libstt? If no i did get through all the steps but ended up with this error , any help would be very appreciated 🙏

function 'virtual Metadata* ModelState::decode_metadata(const DecoderState&, size_t)': native_client/modelstate.cc:73:3: error: too many initializers for 'Metadata' }; ^
``