# Hybrid-AEC-based-on-PercepNet

**Repository Path**: oucxlw/Hybrid-AEC-based-on-PercepNet

## Basic Information

- **Project Name**: Hybrid-AEC-based-on-PercepNet
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 2
- **Forks**: 1
- **Created**: 2021-06-25
- **Last Updated**: 2024-12-08

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Hybrid-AEC-based-on-PercepNet

Hello guys, this is an implementation of "Low-complexity, real-time joint neural echo control and speech enhancement based on PercepNet." published by Dr. Jean-Marc Valin in 2021 AEC challenge, cohosted by Microsoft and IEEE. The paper can be reached by the link below:

https://arxiv.org/abs/2102.05245

This implementation is mainly on Python right now, and the C++ version will go on after sucessful Python implementation.

# Progress List (Python)
- [x] Traditional AEC: Multidelay block frequency (MDF) adaptive filter, Dr. Valin's version
- [x] ERB filterbank generation 
- [x] Pitch tracking
- [x] Comb filter implementation
- [x] Pitch coherence calculation
- [ ] Envelope post-filtering and scale bands
- [ ] Data augmentation and data generation
- [ ] DNN Model training and testing

The pitch coherence calculation has problems, so I took abs value on pitch coherence and alpha to make sure everything works. Not sure if this step is right, if any idea please open an issue on this repository :wink:

# Database
The dataset is open on 2021 AEC Challenge repository:

https://github.com/microsoft/AEC-Challenge

# Renference repository
PercepNet implementation:  https://github.com/jzi040941/PercepNet

MDF filter:                https://github.com/xiph/speexdsp

Pitch tracking algorithm:  https://github.com/dgaspari/pyrapt

ERB filterbank:            https://github.com/wil-j-wil/py_bank