PhaseNet: Discretized Phase Modeling with Deep Neural Networks for Audio Source Separation

By Naoya Takahashi, Purvi Agrawal, Nabarun Goswami, Yuki Mitsufuji. 2018

71 citations recorded by the source corpus.

Previous research on audio source separation based on deep neural networks (DNNs) mainly focuses on estimating the magnitude spectrum of target sources and typically, phase of the mixture signal is combined with the estimated magnitude spectra in an ad-hoc way. Although recovering target phase is assumed to be important for the improvement of separation quality, it can be difﬁcult to handle the periodic nature of the phase with the regression approach. Unwrapping phase is one way to eliminate the phase discontinuity, however, it increases the range of value along with the times of unwrapping, making it difﬁcult for DNNs to model. To overcome this difﬁculty, we propose to treat the phase estimation problem as a classiﬁcation problem by discretizing phase values and assigning class indices to them. Experimental results show that our classiﬁcation-based approach 1) successfully recovers the phase of the target source in the discretized domain, 2) improves signal-to-distortion ratio (SDR) over the regression-based approach in both speech enhancement task and music source separation (MSS) task, and 3) outperforms state-of-the-art MSS.

Subfields: Artificial Intelligence, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Signal Processing

Citations in the corpus

Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesis (2023, background): “Another line of work suggests perceptual phase quantization (Kim, 2003), which has proven promising in deep learning by treating the phase estimation problem as a classification problem (Takahashi et al., 2018).”
The Phasebook: Building Complex Masks via Discrete Representations for Source Separation (2019, unknown): “[18] N. Takahashi, P. Agrawal, N. Goswami, and Y. Mitsufuji, “PhaseNet: Discretized phase modeling with deep neural networks for audio source separation,” Proc.”
Phase-aware Single-stage Speech Denoising and Dereverberation with U-Net (2020, methodology): “Input features were used as a channel-wise concatenation of log-magnitude spectrogram, real and imaginary part of demodulated phase [13], group delay, and delta-phase [27].”
PHASEN: A Phase-and-Harmonics-Aware Speech Enhancement Network (2019, methodology): “There exist some other methods (Takahashi et al. 2018; Takamichi et al. 2018; Masuyama et al. 2019) which process phase reconstruction asynchronously with amplitude estimation.”
D3Net: Densely connected multidilated DenseNet for music source separation (2020, background): “Many types of neural network architecture have been proposed, including feedforward fully connected networks (FNNs) [1, 2], recurrent neural networks (RNNs) [3], convolutional neural networks (CNNs) [4–9] and their combinations [10, 11].”
Phase-aware Speech Enhancement with Deep Complex U-Net (2019, methodology): “Furthermore, the use of real-valued U-Net has been shown to be also effective in many recent audio source separation tasks such as music source separation (Jansson et al., 2017; Stoller et al., 2018; Takahashi et al., 2018b), and speech enhancement (Pascual et al., 2017). Deep Complex U-Net (DCUnet)”