Skip to content

Latest commit

 

History

History
21 lines (16 loc) · 1.46 KB

README.md

File metadata and controls

21 lines (16 loc) · 1.46 KB
Raw

Experiment Code for a Theory of Unsupervised Speech Recognition

This repository contains the source code for the paper "A Theory of Unsupervised Speech Recognition": https://www.researchgate.net/publication/370605684_A_Theory_of_Unsupervised_Speech_Recognition.

Dependencies

How to run it?

  1. Change $CONDA_ROOT, $FAIRSEQ_ROOT, $KALDI_ROOT and $KENLM_ROOT in run_{synthetic, analysis}.sh to those of your own.

  2. For phase transition experiments: bash run_synthetic.sh ${l1,mmd,jsd,wasserstein} ${circulant,debruijn,hypercube}

  3. For further analysis such as the effect of training with discriminator reset, discriminator type and generator type:

bash run_analysis.sh ${gan_type} ${graph_name} ${gen_type} ${discrim_type} 2 2  # The effect of discriminator reset
bash run_analysis.sh ${gan_type} ${graph_name} ${gen_type} ${discrim_type} 4 4  # The effect of discriminator type
bash run_analysis.sh ${gan_type} ${graph_name} ${gen_type} ${discrim_type} 6 6  # The effect of generator type
  1. To generate figures from the paper, please check out synthetic_asr_u.ipynb for more details. Currently, figures for the GAN-based experiments are generated based on manually-created .csv files.