DeepSpeech: DeepSpeech fails on CPUs without AVX support

Hello,

I am running through the standard set-up and was having some trouble on the final step. It appears to be a recurring issue, but the solution I believe is provided hasn’t fixed the problem, which is why I’m assuming there is a bug. Very well may be me though so I’ve tried to describe every step.

I’m using a MacBook Pro 10.11.6 with a 2.4 GHz Intel Core 2 Duo Processor (incredibly old).

Tensorflow has not been installed (instruction don’t seem to indicate it as a prerequisite).

Python is 2.7.

I first cloned the repository, but that doesn’t seem to be important for what’s next. Then I downloaded and unzipped the pre-trained model.

I set up a virtual environment for deepspeech using conda. Then I activated it and ran pip install deepspeech. After installation I ran:

deepspeech location_of_downloaded_pretrained_model.pb location_of_sound.wav

This immediately got the “Illegal instruction: 4” and the code stopped.

After a bit of searching it seems like the problem is AVX2/FMA as documented here (https://github.com/mozilla/DeepSpeech/issues/1057) and here (https://github.com/mozilla/DeepSpeech/issues/1023).

The consensus seemed to be to go to https://tools.taskcluster.net/index/artifacts/project.deepspeech.deepspeech.native_client.master/osx and select deepspeech-0.1.0-cp27-cp27m-macosx_10_10_x86_64.whl (since I’m running python 2.7 with an old Mac).

Then, I uninstalled the current deepspeech with:

pip uninstall deepspeech 

and ran:

pip install deepspeech-0.1.0-cp27-cp27m-macosx_10_10_x86_64.whl 

while in a folder that contained that file.

Seemed to work, but then I ran:

deepspeech location_of_downloaded_pretrained_model.pb location_of_sound.wav

once more and it still gives me the error “Illegal instruction: 4” and the code stopped.

Have I done something wrong here or is this still a bug?