Lost in Transcription, Found in Distribution Shift: Demystifying Hallucination in Speech Foundation Models

TL;DR

This paper investigates hallucination in speech recognition models, revealing how distribution shifts and noise impact hallucination rates, and proposes HER as a new metric to better evaluate transcription reliability in critical applications.

Contribution

It introduces the hallucination error rate (HER) metric and analyzes how model size, architecture, and distribution shifts influence hallucination in ASR systems.

Findings

HER increases with distribution shift ($\alpha=0.91$)

High WER can hide hallucinations, low WER can conceal dangerous errors

Synthetic noise elevates HER significantly

Abstract

Speech foundation models trained at a massive scale, both in terms of model and data size, result in robust systems capable of performing multiple speech tasks, including automatic speech recognition (ASR). These models transcend language and domain barriers, yet effectively measuring their performance remains a challenge. Traditional metrics like word error rate (WER) and character error rate (CER) are commonly used to evaluate ASR performance but often fail to reflect transcription quality in critical contexts, particularly when detecting fabricated outputs. This phenomenon, known as hallucination, is especially concerning in high-stakes domains such as healthcare, legal, and aviation, where errors can have severe consequences. In our work, we address this gap by investigating hallucination in ASR models. We examine how factors such as distribution shifts, model size, and model architecture influence the hallucination error rate (HER), a metric we introduce to quantify hallucinations. Our analysis of over 20 ASR models reveals \numinsights~key insights: (1) High WERs can mask low hallucination rates, while low WERs may conceal dangerous hallucinations. (2) Synthetic noise, both adversarial and common perturbations like white noise, pitch shift, and time stretching, increase HER. (3) Distribution shift correlates strongly with HER ($\alpha = 0.91$). Our findings highlight the importance of incorporating HER alongside traditional metrics like WER to better assess ASR model performance, particularly in high-stakes domains.