Hello everyone,
At work, we are starting an effort to transcribe many things. So far, we’ve been training our favourite cloud service provider and paying for the privilege. So, I thought, if we are getting a good number of people to train the models, let us at least contribute to a cause. I’m really keen into getting it to work, because we would be able to contribute with multiple languages (big company, many countries).
Unfortunately, my workstation is a classic Mac Pro (2x Xeon X5675) and my boss won’t assign me any resources before I have a working concept. I thought of going to Linux, but I can’t do that now, not even dual boot (EFI, can’t see boot screen, can’t see boot drive, a new computer would be nice), but even that wouldn’t help, because I have an AMD GPU. My current OS is the latest macOS (Mac OS) X 10, however Apple calls it these days.
So, after some head-scratching, I successfully compiled and ran v0.5.0-alpha.11-17 without AVX (and, unfortunately, that came as well without SSE4.2), but it runs.
However, I can’t use the pre-trained models to run a POC because of the version mismatch. I downloaded some samples and decided to then run some training but then I get stuck with to run DeepSpeech.py, because it seems to prefer Python3, but then I get to the AVX tensorflow (I compiled everything under Python2, naïve, never used this before, didn’t check the environment), so I can’t get the ds_ctcdecoder to work, it seems to be happy only under Python3, and if I run that, I get the tensorflow that I didn’t compile myself, it tries AVX and boom, it dies.
Any tips?
PS: What a cool project!