LID Models are Actually Accent Classifiers: Implications and Solutions for LID on Accented Speech

TL;DR

This paper reveals that LID models often misclassify accented speech as related languages, analyzes the causes, and proposes input chunking and sequence-level integration methods to improve robustness and accuracy on accented speech.

Contribution

It identifies the accent classification behavior of LID models, analyzes their invariance to speech permutations, and introduces novel methods to improve performance on accented speech without relying on monolingual ASR.

Findings

LID models often confuse accents with related languages.

Input chunking improves model robustness to accents.

Sequence-level information integration reduces accent-language confusion.

Abstract

Prior research indicates that LID model performance significantly declines on accented speech; however, the specific causes, extent, and characterization of these errors remain under-explored. (i) We identify a common failure mode on accented speech whereby LID systems often misclassify L2 accented speech as the speaker's native language or a related language. (ii) We present evidence suggesting that state-of-the-art models are invariant to permutations of short spans of speech, implying they classify on the basis of short phonotactic features indicative of accent rather than language. Our analysis reveals a simple method to enhance model robustness to accents through input chunking. (iii) We present an approach that integrates sequence-level information into our model without relying on monolingual ASR systems; this reduces accent-language confusion and significantly enhances performance on accented speech while maintaining comparable results on standard LID.