IBM Estimated Using Deep Xi

Deep Xi from [1] is now used instead of the bidirectional recurrent neural network (BRNN) from [2]. Deep Xi is a deep learning approach to a priori SNR estimation, implemented in TensorFlow. The a priori SNR estimated by Deep Xi is used to compute an ideal binary mask (IBM) estimate.

Deep Xi can be found here.

References

https://doi.org/10.1016/j.specom.2019.06.002

[1] A. Nicolson and K. K. Paliwal, "Deep Learning For Minimum Mean-Square Error Approaches to Speech Enhancement", Speech Communication, 2019, ISSN 0167-6393, https://doi.org/10.1016/j.specom.2019.06.002.

[2] Nicolson, A. and Paliwal, K.K., 2018. Bidirectional Long-Short Term Memory Network-based Estimation of Reliable Spectral Component Locations. Proc. Interspeech 2018, pp.1606-1610