AI powered speech denoising and enhancement
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
2026-10-05 06:48:28 -04:00
config Finalize 0.0.1 2023-12-14 15:45:06 +08:00
resemble_enhance Fix custom model directory resolution 2026-10-05 06:44:08 -04:00
tests Fix custom model directory resolution 2026-10-05 06:44:08 -04:00
.gitignore Get rid of git lfs & update README 2024-01-04 20:23:03 +08:00
app.py Init 2023-12-13 00:03:07 +08:00
LICENSE Init 2023-12-13 00:03:07 +08:00
packages.txt Init 2023-12-13 00:03:07 +08:00
pyproject.toml Init 2023-12-13 00:03:07 +08:00
README.md Fix custom model directory resolution 2026-10-05 06:44:08 -04:00
requirements.txt Init 2023-12-13 00:03:07 +08:00
setup.py Switch to cpu if cuda is not available 2023-12-15 11:06:27 +08:00

Resemble Enhance

PyPI Hugging Face Space License Webpage

https://github.com/resemble-ai/resemble-enhance/assets/660224/bc3ec943-e795-4646-b119-cce327c810f1

Resemble Enhance is an AI-powered tool that aims to improve the overall quality of speech by performing denoising and enhancement. It consists of two modules: a denoiser, which separates speech from a noisy audio, and an enhancer, which further boosts the perceptual audio quality by restoring audio distortions and extending the audio bandwidth. The two models are trained on high-quality 44.1kHz speech data that guarantees the enhancement of your speech with high quality.

Usage

Installation

Install the stable version:

pip install resemble-enhance --upgrade

Or try the latest pre-release version:

pip install resemble-enhance --upgrade --pre

Enhance

resemble-enhance in_dir out_dir

Denoise only

resemble-enhance in_dir out_dir --denoise_only

Web Demo

We provide a web demo built with Gradio, you can try it out here, or also run it locally:

python app.py

Train your own model

Data Preparation

You need to prepare a foreground speech dataset and a background non-speech dataset. In addition, you need to prepare a RIR dataset (examples).

data
├── fg
│   ├── 00001.wav
│   └── ...
├── bg
│   ├── 00001.wav
│   └── ...
└── rir
    ├── 00001.npy
    └── ...

Training

Denoiser Warmup

Though the denoiser is trained jointly with the enhancer, it is recommended for a warmup training first.

python -m resemble_enhance.denoiser.train --yaml config/denoiser.yaml runs/denoiser

Enhancer

Then, you can train the enhancer in two stages. The first stage is to train the autoencoder and vocoder. And the second stage is to train the latent conditional flow matching (CFM) model.

Stage 1
python -m resemble_enhance.enhancer.train --yaml config/enhancer_stage1.yaml runs/enhancer_stage1
Stage 2
python -m resemble_enhance.enhancer.train --yaml config/enhancer_stage2.yaml runs/enhancer_stage2

Blog

Learn more on our website!