Skip to content

Latest commit

 

History

History
41 lines (27 loc) · 1.77 KB

description.md

File metadata and controls

41 lines (27 loc) · 1.77 KB
Raw

TAK1

Naoya Takahashi¹, Nabarun Goswami², Yuki Mitsufuji¹

¹Sony Corporation, Audio Technology Development Department, Tokyo, Japan
²Sony India Software Center, Bangalore, India

Naoya.Takahashi [at] sony.com

Additional Info

  • is_blind: no
  • additional_training_data: no

Supplemental Material

  • Code: not available
  • Demos: not available

Method

This submission uses a multi-scale multi-band DenseLSTM (MMDeseLSTM), which is an extetnsion of MMDenseNet [1]. In MMDenseLSTM, some of dense-blocks, typically in low scale, equip a LSTM layer, as shown in Figure 1. Each LSTM layer is one layer bi-directional LSTM with maximum 128 cells. For each instrument, a MMDeseLSTM is trained to predict the target instrument amplitude from the mixture amplitude in the STFT domain (frame size: 4096, hop size: 1024). The raw output of each network is then combined by a multichannel Wiener filter as described in [2] where we estimate the power spectral densities and spatial covariance matrices from the DNN outputs.
More details, experiments and analysis is described in a paper which is submitted to elsewhere.

The network is trained on train part of musdb for 60 epochs, where a training curve saturate.

Figure 1. MDenseLSTM architecture (scale=3). A LSTM layer is added to the bottleneck DenseBlock of MDenseNet[1].

References

  1. N. Takahashi and Y. Mitsufuji: Multi-scale multi-band DenseNets for audio source separation, Proc. WASPAA, 2017
  2. A. A. Nugraha, A. Liutkus, and E. Vincent. "Multichannel music separation with deep neural networks." EUSIPCO, 2016.
  3. N. Takahashi, N. Goswami and Y. Mitsufuji: MMDenseLSTM: an efficient two way modeling for audio source separation, arXiv pre-print, 2018