/UNet-MISO

unofficial implementation of "A Causal U-net based Neural Beamforming Network for Real-Time Multi-Channel Speech Enhancement"

Primary LanguagePython

UNet-MISO

An end-to-end lightweight multichannel speech enhancement network

A minimum implementation of A Causal U-net based Neural Beamforming Network for Real-Time Multi-Channel Speech Enhancement -- Interspeech 2021


This model acts as a weighted and sum beamformer

How to use:

1 add unet_miso.py to your model directory
2 import UNet as model and ready to go 👻

Model params:

- params: 1.1M
- n_fft: number of sample points for STFT
- hop_length: hop size for STFT
- mics: number of channels

Metrics

Speech and noise dataset from DNS 2022 challenge, RIR is simulated for eight-microphone circular array

                  sisnr(dB)    snr(dB)    stoi
1. noisy.          7.588       11.499     0.772
2. enhanced       12.839       14.039     0.837

Samples

8-channel noisy wav

target single-channel wav

enhanced single-channel wav