# Learning Similarity Functions for Pronunciation Variations

## TL;DR

This paper introduces neural network-based methods to learn pronunciation similarity functions, improving lexical access in ASR systems by handling pronunciation variations more effectively.

## Contribution

It proposes two novel RNN-based approaches for learning pronunciation similarity, enhancing lexical access and error prediction in speech recognition.

## Key findings

- RNN-based methods outperform Bayesian approaches in lexical access tasks.
- The methods effectively handle pronunciation variations in spontaneous speech.
- Results demonstrate improved accuracy in ASR error prediction.

## Abstract

A significant source of errors in Automatic Speech Recognition (ASR) systems is due to pronunciation variations which occur in spontaneous and conversational speech. Usually ASR systems use a finite lexicon that provides one or more pronunciations for each word. In this paper, we focus on learning a similarity function between two pronunciations. The pronunciations can be the canonical and the surface pronunciations of the same word or they can be two surface pronunciations of different words. This task generalizes problems such as lexical access (the problem of learning the mapping between words and their possible pronunciations), and defining word neighborhoods. It can also be used to dynamically increase the size of the pronunciation lexicon, or in predicting ASR errors. We propose two methods, which are based on recurrent neural networks, to learn the similarity function. The first is based on binary classification, and the second is based on learning the ranking of the pronunciations. We demonstrate the efficiency of our approach on the task of lexical access using a subset of the Switchboard conversational speech corpus. Results suggest that on this task our methods are superior to previous methods which are based on graphical Bayesian methods.