How to compile the native client's ctcdecoder?

[heiner_schroeder]

Hi,
im trying to make changes to the native client. For the time being just to understand how everything exactly works and probably later to implement some stuff myself. If i make changes to the C+±files in the native_client source files and compile it via ‘make deepspeech’ my changes affect the output as expected.

However, if I try to make changes in the ctcdecode subdirectory I dont get any effect on the output (even though it should). I tried to compile it via ‘make’ within the ctcdecode subdirectory. Now, something is compiling, but it has no effect on the deepspeech binary.

I guess its obvious that im not an expert on makefiles and everything related. So, how do I compile it right so that changes I make in ctcdecode subdirectory affect the output of the deepspeech binary?

[lissyx ((slow to reply) [NOT PROVIDING SUPPORT])]

I tried to compile it via ‘make’ within the ctcdecode subdirectory. Now, something is compiling, but it has no effect on the deepspeech binary.

Have you installed the python package after rebuilding it ?

Why do you need to rebuild ctcdecoder, what kind of changes are you making ?

[heiner_schroeder]

Correct me, if im wrong. But if I just use the deepspeech binary in the native_client directory, there is no need to install anything python related, is it?

And yes, I know that I’m using the C++ -binary via

./native_client/deepspeech --model ....

Manipulating and compiling the source files in the native_client directory does have an effect on the output, however not the changes I make in the ctcdecode subdirectory.

I try to understand what is happening at every stage, since I might want to make changes. But first I need to understand what is happening exactly and being able to output whatever I want, would be quite helpfull.

And since the native_client code uses functions from ctcdecode I need to compile everything, not just the native_client directory.

[lissyx ((slow to reply) [NOT PROVIDING SUPPORT])]

But if I just use the deepspeech binary in the native_client directory, there is no need to install anything python related, is it?

That’s right. But you mention rebuilding ctcdecoder, and running make in native_client/ctcdecode which is explicitely meant for rebuilding the ds_ctcdecoder Python package required at training time.

Manipulating and compiling the source files in the native_client directory does have an effect on the output, however not the changes I make in the ctcdecode subdirectory.

Can you be more specific ? What changes are you doing ?

And since the native_client code uses functions from ctcdecode I need to compile everything, not just the native_client directory.

Have you carefully read the documentation ? Have you rebuilt libdeepspeech.so as well ?