Building Coqui with TFLite delegate support #1925
Unanswered
jeremiahrose
asked this question in
Q&A
Replies: 1 comment 1 reply
-
All you need to do is set the environment variable and verify the model is behaving as intended. No need to change the build (or even build yourself).
Yep. Code here: STT/native_client/tflitemodelstate.cc Lines 102 to 156 in df26eca |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi there!
I gather from the docs and some old Deepspeech discussions that building and using Coqui STT with the GPU, NNAPI and Hexagon delegates was still in the experimental phase around 1 year ago (early 2020).
I'm wondering where that is at now, and what I would need to do to get it working?
Specifically I want to know the proper way to:
libstt.so
,stt
native client, with delegates enabled. What Bazel flags are required? (I'm cross-compiling to ARMv7 Linux)libstt.so
?stt
native client with a delegate (is e.gSTT_TFLITE_DELEGATE=gpu
all that's required?)Happy to do whatever debugging is required to get this to work. Also hoping that this thread might become a starting point for other people who are trying to get this working.
Beta Was this translation helpful? Give feedback.
All reactions