TorchBits
Deep signal processing for audio and images.
- Role
- Creator
- Type
- Solo
- Year
- 2024
- Status
- Open source
- Category
- Languages and Libraries
- Main stack
- Python, C++ +1
Overview
TorchBits is a minimalist library for deep signal processing. It uses low level C++ with Python bindings for fast matrix multiplication, includes basic convolutional network building blocks inspired by ResNet, and offers signal tools for spatial and frequency analysis.
It was designed to be memory efficient so it can run on edge devices and low spec machines, with optional GPU support through CuPy.
Highlights
- •Custom C++ matrix operations with Python bindings.
- •CNN building blocks for denoising and classification.
- •Low memory footprint for edge devices.
Built with
- Python
- C++
- NumPy
