Wind Noise Reduction with a Diffusion-based Stochastic Regeneration Model
This website contains supplementary material to the paper:
- StoRM: A Diffusion-based Stochastic Regeneration Model for Speech Enhancement and Dereverberation) and data generation scripts is available at https://github.com/sp-uhh/storm
Simulated dataset
440o030443snr-14.wav 447c0201570snr70.wav 441c020483snr53.wav 442c020k181snr-16.wav Input SNR -1.4 dB 7.0 dB 5.3 dB -1.6 dB Clean Noisy FCN+SANM [2] DBLSTM-U [3] NCSN++M [4] SGMSE+M [4, 5] StoRM-G [1, 6] Unseen dataset with real-recorded wind noise
abwindblocksn0mod.wav abwindblocksn-5mod.wav abwindblocksn-10unmod.wav Input SNR 0.0 dB -5.0 dB -10.0 dB Noisy FCN+SANM [2] DBLSTM-U [3] NCSN++M [4] SGMSE+M [4, 5] StoRM-G [1, 6] References
[1] J-M. Lemercier, J. Thiemann, R. Koning and T. Gerkmann. Wind Noise Reduction with a Diffusion-based Stochastic Regeneration Model. arXiv preprint arXiv:2306.12867. 2023.
[2] H. Bai, F. Ge, and Y. Yan, DNN-based speech enhancement using soft audible noise masking for wind noise reduction. China Communications, vol. 15, no. 9, pp. 235-243. 2018.
[3] J. Lee, K. Kim, T. Z. Shabestary, and H.-G. Kang. Deep bi-directional long short-term memory based speech enhancement for wind noise reduction. Hands-free Speech Communications and Microphone Arrays (HSCMA). 2017.
[4] J-M. Lemercier, J. Richter, S. Welker and T. Gerkmann. Analysing Diffusion-based Generative Approaches versus Discriminative Approaches for Speech Restoration. ICASSP. 2023.
[5] J. Richter, S. Welker, J-M. Lemercier, B. Lay, and T. Gerkmann. Speech Enhancement and Dereverberation with Diffusion-Based Generative Models. IEEE TASLP. 2023.
[6] J-M. Lemercier, J. Richter, S. Welker and T. Gerkmann. StoRM: A Diffusion-based Stochastic Regeneration Model for Speech Enhancement and Dereverberation. arXiv preprint arXiv:2212.11851. 2022.