Automate concurrent whisper transcription in AWS spot machines for large sets of audio files

[pulijon]

This project is aimed at transcribing large amounts of audio (in my case, collections of podcasts). It includes an automation system based on Terraform and Ansible, which allows the use of AWS machines with sufficient CPU and GPU to run several transcription processes concurrently. In this way, it is not necessary to have a machine with GPU acceleration to use it.

The following table shows details of cost and speed, compared to the Amazon transcription service.