Neural networks are increasingly successful for audio applications such as source separation, automatic speech recognition, and sequence embedding. Yet the phase spectra of audio signals are usually not modeled directly. We claim that, for many applications, it may be advantageous to jointly model phase and magnitude spectra, whether from Fourier-transformed or from raw waveforms.
The ongoing project seeks to describe basic properties of complex-valued neural networks when applied to machine learning tasks on audio. We recently presented a poster on preliminary ideas at the 2015 Speech and Audio in the Northeast (SANE) workshop. A paper on the mathematical background for gradient based learning on complex valued nets is in submission to the 2016 International Conference on Learning Representations (Sarroff, Shepardson, & Casey, 2015).
Sarroff, A. M., Shepardson, V., & Casey, M. A. (2015). Learning Representations Using Complex-Valued Nets. ArXiv e-Prints. Submitted to the 2016 International Conference on Learning Representations (ICLR).
Neural networks are increasingly successful for audio applications such as source separation, automatic speech recognition, and sequence embedding. Yet the phase spectra of audio signals are usually not modeled directly. We claim that, for many applications, it may be advantageous to jointly model phase and magnitude spectra, whether from Fourier-transformed or from raw waveforms.
The ongoing project seeks to describe basic properties of complex-valued neural networks when applied to machine learning tasks on audio. We recently presented a poster on preliminary ideas at the 2015 Speech and Audio in the Northeast (SANE) workshop. A paper on the mathematical background for gradient based learning on complex valued nets is in submission to the 2016 International Conference on Learning Representations (Sarroff, Shepardson, & Casey, 2015).
Sarroff, A. M., Shepardson, V., & Casey, M. A. (2015). Learning Representations Using Complex-Valued Nets. ArXiv e-Prints. Submitted to the 2016 International Conference on Learning Representations (ICLR).
http://www.cs.dartmouth.edu/~sarroff/projects/cvnn/
https://arxiv.org/pdf/1511.06351v1.pdf