WARNING: this repository may become private in a near future.
This project started as a technical test in an interview process. It aims to compare the performances of two Whisper implementations: whisper-burn and whisper-rs.
The requirements on Ubuntu include:
sudo apt install libclang-dev # required to compile the whisper-rs crate
sudo apt install ffmpeg # called by whisper_cpp_wrapper::ffmpeg_decoder::use_ffmpegwhisper-rs works while whisper-burn, with the last commit of 2023-08-07, does not output what is expected.
To execute the code, launch a phony target from the Makefile. It will download the needed files before the execution.
For example, make bench measures the performance of transcripting an audio file with
whisper-rs with several models.