FAT (A fast audio transformer library)

FAT is a collection of feature extraction algorithms such as Filter Banks, MFCCs, MFSCs, etc…

FAT is written in Rust & has bindings for Python. The motivation behind FAT is to have a single library or crate for training (done in Python) & deployment (done in Rust) of any ASR models that I develop for Osmos or anything else.

Features:

• Filter Banks
• MFCC
• MFSC
• Rust Python bindings

---

Installation & Building Fat

Create python enviornment

python3 -m venv .env
source .env/bin/activate

Install requirements

pip install -r requirements.txt

Generate python wheel for FAT

maturin develop --release
maturin build

---

These are some great resources to learn more about FFTs, FilterBanks, MFCCs, etc..

• https://www.inf.ed.ac.uk/teaching/courses/asr/2018-19/asr02-signal-handout.pdf
• https://haythamfayek.com/2016/04/21/speech-processing-for-machine-learning.html
• https://www.youtube.com/watch?v=spUNpyF58BY
• https://tspace.library.utoronto.ca/bitstream/1807/44123/1/Mohamed_Abdel-rahman_201406_PhD_thesis.pdf