Voice check-in system based on unsupervised learning

Abstract

Aiming at the problem of low recognition rate in Multi-object recognition, a speech check-in system based on unsupervised learning was proposed from the perspective of feature extraction and utilization of tag missing data. In this method, the Restricted Boltzmann Machine was used to extract the feature parameters and the Hidden Markov Model model was used to train the speech data. Deep Belief Network provided observation probability of original data, and Hidden Markov Model obtained likelihood probability of data through forward algorithm, which served as the basis of speech recognition. Experiments showed that when the number of hidden nodes was 30, in the Restricted Boltzmann Machine network, the error of network reconstruction was less than 20%, and the recognition ability of the speech check-in system was improved, which was of great significance for the speech recognition problem of multiple recognition targets.