Abstract
Noise comes from a variety of sources in real world, which makes a lot of non-stationary noises, and it is difficult to find target speech from noisy auditory signals. Recently, adversarial learning models get attention for its high performance in the field of noise control, but it has limitation to depend on the one-to-one mapping between the noisy and the target signals, and unstable training process due to the various distributions of noise. In this paper, we propose a novel deep learning model to learn the noise and target speech distributions at the same time for improving the performance of noise cancellation. It is composed of two generators to stabilize the training process and two discriminators to optimize the distributions of noise and target speech, respectively. It helps to compress the distribution over the latent space, because two distributions from the same source are used simultaneously during adversarial learning. For the stable learning, one generator is pre-trained with minimum sample and guides the other generator, so that it can prevent mode collapsing problem by using prior knowledge. Experiments with the noise speech dataset composed of 30 speakers and 90 types of noise are conducted with scale-invariant source-to-noise ratio (SI-SNR) metric. The proposed model shows the enhanced performance of 7.36, which is 2.13 times better than the state-of-the-art model. Additional experiment on −10, −5, 0, 5, and 10 dB of the noise confirms the robustness of the proposed model.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
References
Sharma, M.K., Vig, R.: Ambulance siren noise reduction using LMS and FXLMS algorithms. Indian J. Sci. Technol. 9(47), 1–6 (2016)
Cohen, I.: Multichannel Post-filtering in nonstationary noise environments. IEEE Trans. Signal Process. 52(5), 1149–1160 (2004)
Pascual, S., Bonafonte, A., Serra, J.: SEGAN: speech enhancement generative adversarial network arXiv:1703.09452 (2017)
Salimans, T., Goodfellow, I., Zaremba, W., Cheung, V., Radford, A., Chen, X.: Improved techniques for training Gans. In: Neural Information Processing Systems, pp. 2234–2242 (2016)
Mahal, H.N., Mudge, P., Nandi, K.A.: Noise removal using adaptive filtering for ultrasonic guided wave testing of pipelines. In: Annual Conference of the British Institute of Non-Destructive Testing, pp. 19–27 (2019)
Tamura, S., Waibel, A.: Noise reduction using connectionist models. In: IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 553–556 (1988)
Lin, J., et al.: Speech enhancement using forked generative adversarial networks with spectral subtraction. In: Interspeech, pp. 3163–3167 (2019)
Lim, K.-H., Kim, J.-Y., Cho, S.-B.: Non-stationary noise cancellation using deep autoencoder based on adversarial learning. In: Yin, H., Camacho, D., Tino, P., Tallón-Ballesteros, A.J., Menezes, R., Allmendinger, R. (eds.) IDEAL 2019. LNCS, vol. 11871, pp. 367–374. Springer, Cham (2019). https://doi.org/10.1007/978-3-030-33607-3_40
Nguyen, T., Le, T., Vu, H., Phung, D.: Dual discriminator generative adversarial nets. In: Neural Information Processing Systems, pp. 2670–2680 (2017)
Hoang, Q., Nguyen, T.D., Le, T., Phung, D.: MGAN: training generative adversarial nets with multiple generators. In: International Conference on Learning Representation, pp. 1–24, 2018
Kim, J.Y., Bu, S.J., Cho, S.B.: Hybrid deep learning based on GAN for classifying BSR noises from invehicle sensors. In: de Cos Juez, F., et al. (eds.) Hybrid Artificial Intelligent Systems, vol. 10870, pp. 27–38. Springer, Heidelberg (2018). https://doi.org/10.1007/978-3-319-92639-1_3
Kim, J.Y., Bu, S.J., Cho, S.B.: Zero-day malware detection using transferred generative adversarial networks based on deep autoencoders. Inf. Sci. 460–461, 83–102 (2018)
Luke, M., Ben, P., David, P., Jascha, S.D.: Unrolled generative adversarial networks. arXiv:1611.02163 (2016)
He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: IEEE Conference on Computer Vision and Pattern Recognition, pp. 770–778 (2016)
Doersch, C.: Tutorial on variational autoencoders arXiv:1606.05908 (2016)
Valentini, C.: Noisy speech database for training speech enhancement algorithms and TTS Models. University of Edinburgh. School of Informatics. Centre for Speech Research (2016)
Hu, G., Wang, D.L.: A tandem algorithm for pitch estimation and voiced speech segregation. IEEE Trans. Audio Speech Lang. Process. 18, 2067–2079 (2010)
Luo, Y., Mesgarani, N.: TasNet: surpassing ideal time-frequency masking for speech separation arXiv:1809.07454 (2018)
Acknowledgement
This work was supported by Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korean government (MSIT) (No. 2020-0-01361, Artificial Intelligence Graduate School Program (Yonsei University)) and grant funded by 2019 IT promotion fund (Development of AI based Precision Medicine Emergency System) of the Korean government (MSIT).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Lim, KH., Kim, JY., Cho, SB. (2020). Generative Adversarial Network with Guided Generator for Non-stationary Noise Cancelation. In: de la Cal, E.A., Villar Flecha, J.R., Quintián, H., Corchado, E. (eds) Hybrid Artificial Intelligent Systems. HAIS 2020. Lecture Notes in Computer Science(), vol 12344. Springer, Cham. https://doi.org/10.1007/978-3-030-61705-9_1
Download citation
DOI: https://doi.org/10.1007/978-3-030-61705-9_1
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-61704-2
Online ISBN: 978-3-030-61705-9
eBook Packages: Computer ScienceComputer Science (R0)