Hello, I have followed the documentation in native_client to build the libdeepspeech.so library to be used in an c++ application.
I am able to use the DS functions and its internal model, however I just need the ctcdecoder to decode my own NN outputs and I can not access them.
I see there is a way for doing this in Python by building the ds_ctcdecoder package. Is there a way to do something similar in c++ or to directly access the functions in ctcdecode?
Thanks
lissyx
((slow to reply) [NOT PROVIDING SUPPORT])
2
Are you aware we ship those prebuilt ? Is there any reason you need to rebuild ?
I’m not sure to follow you here. The decoder is already plugged and used by libdeepspeech.so.
Are you trying to re-use ds_ctcdecoder for something else ?
Because this is not part of the public API we expose, so yes, it’s not directly accessible.
Thanks for the reply, yes I am aware about the prebuilt library and I also have this.
And yes, I am trying to use just the ds_ctcdecoder with my own predicted outputs, but in c++ not in Python. Since it is not directly accessible from the API , is there a way to build just the decoder ?
lissyx
((slow to reply) [NOT PROVIDING SUPPORT])
4
Can you elaborate on what you are doing exactly ?
Yes, you need to dig into the native_client/BUILD file to see the targets, especially decoder one. Now, this is Bazel, but it should give you a good insight.