Topological analysis of unaligned audio and text data

We have performed preliminary work on topological analysis of audio and text data for unsupervised speech processing. The assumes that phoneme frequencies contextual relationships are similar in the acoustic domains same language. Accordingly, this allowed creation a mapping between these spaces considers their geometric structure. As first step, generative methods based variational autoencoders were chosen to map into two latent vector spaces. In next stage, persistent homology used analyze structure Although results obtained support idea similarity spaces, further research is needed correctly as well evaluate real effect including information autoencoder training process.