作者: Kelan Kuang
Email: kuangkelan@mail.ioa.ac.cn
Multi-Channel Speech Enhancement results
This repo shows enhanced speech of TriU-Net and other models.
Real recordings
SNR: -5 dB
Spk 1 is a 24-year-old male, while spk 2 is a 24-year-old female. Both spk 1 and spk 2 speek in Mandarin.
| Signals | Spk 1 babble | Spk 2 babble | Spk 1 f16 | Spk 2 f16 |
|---|---|---|---|---|
| noisy |  |
 |
|
 |
| clean |  |
 |
|
|
| oracl MB-MVDR |  |
 |
 |
 |
| EaBNet |  |
 |
 |
 |
| FaSNet |  |
 |
 |
 |
| Model A |  |
 |
 |
 |
| TriU-Net |  |
 |
 |
 |
Synthetic Data
| Signals | Spk 1 SNR:-5.36 dB | Spk 2 SNR:-3.43 dB |
|---|---|---|
| noisy |  |
 |
| clean |  |
 |
| EaBNet |  |
 |
| FaSNet |  |
 |
| Model A |  |
 |
| TriU-Net |  |
 |
Other samples
Other SNR scenarios See https://github.com/CaA23187/TriU-Net_demo