Flow-based Speech Analysis

From cslt Wiki
Jump to: navigation, search


  • We present a preliminary investigation on unsupervised speech factorization based on the normalization flow model.

This model constructs a complex invertible transform, by which we can project speech segments into a latent code space where the distribution is a simple diagonal Gaussian.

  • Our preliminary investigation on the TIMIT database shows that this code space exhibits favorable properties such as denseness and pseudo linearity,

and perceptually important factors such as phonetic content and speaker trait can be represented as particular directions within the code space.

  • Index Terms: speech factorization, normalization flow, deep learning


  • Dong Wang, Haoran Sun, Yunqi Cai, Lantian Li


  • Haoran Sun, Yunqi Cai, Lantian Li, Dong Wang, "On Investigation of Unsupervised Speech Factorization Based in Normalization Flow", 2019. pdf

Source Code

  • Glow model implemented by Yuki-Chai in PyTorch. glow-pytorch


  • Right click "figs" or "wavs" and select "save as" to save the spectrogram figures or related audio files.

Fig3 flow.jpg

  • Interpolation: figs wavs (right click)

Fig4 flow.jpg

Fig5 flow.jpg

Future Work

  • To conduct more thorough studies on large databases and continuous speech.
  • To investigate discriminative flow models which take class information into consideration.