Preliminary tasks of unsupervised speech recognition based on unaligned audio and text data

Zhanibek Kozhirbayev, Talgat Islamgozhayev, Zhandos Yessenbayev, Altynbek Sharipbay

2022

DOI 10.1109/icemis56295.2022.9914249

Data confidence 70 medium

How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

We present herein our work on the preliminary tasks of unsupervised speech recognition using only unaligned audio and text datasets. The motivation for this is the general progress in generative models. Using the assumption that the frequencies and contextual relationships of words are close in audio and text domains for the same language. The experiments on acoustic and test data using the variational autoencoder (VAE) architecture were conducted on word level. Our plan to extract the encoding part of the acoustic VAE and the decoding part of the text VAE to build a joint VAE.