#speech enhancement

Resemble Enhance is an AI tool designed for improving speech quality through both denoising and enhancement. It separates speech from background noise and enhances audio quality by correcting distortions. The tool is trained with high-quality 44.1kHz data to ensure excellent results and provides easy installation options for stable and pre-release versions. A web demo via Gradio offers practical experience. Custom model training is supported, allowing flexible audio processing solutions.

This document presents a noise suppression library that uses recurrent neural networks for real-time full-band speech enhancement. It supports AVX2 architectures and includes a tool for raw PCM file processing. Pre-training, the library leverages openly available datasets of clean speech and noise data, utilizing techniques such as data augmentation. Training is streamlined via a Python script that prepares features and converts models to C files. Loadable models are scalable to accommodate various deployment needs.