A Modified Algorithm for Multiple Input Spectrogram Inversion

Dongxiao Wang, Hirokazu Kameoka, Koichi Shinoda


We propose a new algorithm to estimate the phase of speech signal in the mixture of audio sources under the assumption that the magnitude spectrum of each source is given. The previous method, multiple input spectrogram inversion algorithm (MISI), often performs poorly when the magnitude spectrograms estimated are not accurate. This may be because it imposes a strict constraint that the summation of source waveforms should be exactly the same as the mixture waveform. Our proposing algorithm employs a new objective function in which this constraint is relaxed. In this objective function, the difference between the summation of source waveforms and the mixture waveform is the target to be minimized. The performance of our method, modified MISI is evaluated on two different experimental settings. In both settings it improves the audio source separation performance compared to MISI.


 DOI: 10.21437/Interspeech.2019-3242

Cite as: Wang, D., Kameoka, H., Shinoda, K. (2019) A Modified Algorithm for Multiple Input Spectrogram Inversion. Proc. Interspeech 2019, 4569-4573, DOI: 10.21437/Interspeech.2019-3242.


@inproceedings{Wang2019,
  author={Dongxiao Wang and Hirokazu Kameoka and Koichi Shinoda},
  title={{A Modified Algorithm for Multiple Input Spectrogram Inversion}},
  year=2019,
  booktitle={Proc. Interspeech 2019},
  pages={4569--4573},
  doi={10.21437/Interspeech.2019-3242},
  url={http://dx.doi.org/10.21437/Interspeech.2019-3242}
}