A fool-proof guide to succesfully installing ctcdecode on Docker, Google Colab or your own PC

[RafaelAmauri]

This is a fool-proof guide to installing ctcdecode.

I have repeated these steps five times across four different servers running different versions of Linux. I have tested these steps on Ubuntu 20.04, Ubuntu 22.04 and Arch Linux.

The first thing you must have installed on your environment is gcc and g++. ctcdecode is NOT fully written in Python, it is instead a mix of Python and C, and uses gcc/g++ to compile the C code in a way that Python can understand. Because of this, make sure you have gcc, g++ and the libs for compiling code installed on your system.

For Ubuntu, run:

apt update -y
apt install gcc g++ build-essential

Now, install conda (pip unfortunately did not work for me no matter how much I tried using it. Conda worked every time I repeated the installation process) and then run:

conda create -n [your environment's name]
conda install -n [your environment's name] -y pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia
conda install -n [your environment's name] python=3.9

Now, run enter your new conda environment with

conda activate [your environment's name]

and run

python3 -c "import torch;print(torch.__version__)"

You should get 1.13.1 as a result.

If you got this far, you should be ready to install ctcdecode. To do that, simply run

cd ctcdecode; pip install .

And it should install without problems.

You can verify that it was successfully installed with

python3 -c "import ctcdecode;ctc = ctcdecode.CTCBeamDecoder(\"test\", beam_width=10, blank_id=\"test\");print(ctc)"

If you got <ctcdecode.CTCBeamDecoder object at 0x7fdadc664f10> it means that ctc is successfully installed!