Code for paper "NOISE-ROBUST SPEECH RECOGNITION WITH 10 MINUTES UNPARALLELED IN-DOMAIN DATA".
Simu-GAN: The GAN model to generated simulated noisy speech from clean speech.
Dual-path ASR: The augmented ASR model that is trained by simulated noisy speech and clean speech.
The simulated noisy speech and its corresponding clean speech and real noisy speech are available at: https://chrisole.github.io/ICASSP2022-demo/