孙浩然

Haoran Sun


Center for Speech and Language Technologies, and
Department of Computer Science and Technology,
Tsinghua University, Beijing, China

I am studying for my master's degree in CSLT, and Department of Computer Science and Technology, Tsinghua University,
and working on Deep learning in Speech factorization. I am supervised by Prof. Thomas Fang Zheng and Prof. Dong Wang.

General Info

  • Date of BirthJul 13, 1996
  • AddressFit Building 1-303, Tsinghua University, Beijing, China
  • E-mailshr19@mails.tsinghua.edu.cn
  • Tel. +86 17 8888 42679

Research Interest

    ●     Speech Processing

    ●     Speech Factorization

    ●     Deep Generative models

Skills

  • Python
  • Shell
  • C/C++
  • Matlab

  • Pytorch
  • Tensorflow
  • Kaldi

Curriculum Vitae





Education

Researches

  • 2021

    How Speech is Recognized to Be Emotional - A Study Based on Information Decomposition(APSIPA2021)

    PDF

    We present a preliminary study on various emotional factors and investigate how each of them impacts modern emotion recognition systems. Our results show that rhythm is the most important component for emotional expression. Moreover, the cross-corpus results are very bad (even worse than guess), demonstrating that the present speech emotion recognition model is rather weak. While by removing one or several unimportant components, the cross-corpus results can be improved.

  • 2020

    Deep Generative Factorization For Speech Signal

    PDF WEB

    We presented a speech factorization approach based on a novel factorial discriminative normalization flow model (factorial DNF). Experiments conducted on a two-factor case that involved phonetic contents and speaker traits demonstrated that the proposed factorial DNF had powerful capability to factorize speech signals and outperformed several comparative models in terms of information representation and manipulation.

  • 2019

    On Investigation of Unsupervised Speech Factorization Based on Normalization Flow

    PDF WEB

    we presented a preliminary investigation on unsupervised speech factorization based on the normalization flow model. This model constructed a complex invertible transform, by which we could project speech segments into a latent code space where the distribution was a simple diagonal Gaussian. Our preliminary investigation on the TIMIT database showed that this code space exhibited favorable properties such as denseness and pseudo linearity, and perceptually important factors such as phonetic contents and speaker traits could be represented as particular directions within the code space.

My Supervisors

My Colleagues