How To Run Meta’s New Model SeamlessM4T on MacOS

Meta’s new Speech/Text translation model. Image Credit: Meta SeamlessM4T team

Meta released their new model for speech/text translation and transcription called SeamlessM4T. Here is how you can run the model locally on your MacOS.

Running the model on MacOS takes work; it requires several complicated steps to get things to work correctly. I spent a few hours getting it to work, downloading the correct packages, building C++ extensions, and running everything.

Getting Familiar With All the Requirements

The code for the model is in this GitHub Repo.

But if you follow the README.md on your local Macbook, it won’t work as one of the dependencies, fairseq2, also from Meta, is not available for MacOS currently. So it needs to be built.

The code for this particular dependency is in this Github Repo.

To get this to build on MacOS requires the most amount of effort.

Apart from this, I am guessing most Mac users use Homebrew to install all other dependencies; if not, you can get it at this link.

We will need Homebrew for one dependency called libsndfile to interact with WAV files. Here’s the link.

You will also need some version of Python installed. I am using Python 3.11, but the examples show you can use older…