Whisper Hallucination Detection and Mitigation via Hidden Representation Steering and Sparse AutoEncoders

2026-06-05

AI Summary

This work addresses the tendency of Whisper speech recognition models to generate fluent yet erroneous “hallucinated” transcriptions on non-speech audio. The study reveals, for the first time, that hallucination signals are linearly separable within the encoder’s hidden representations and predominantly localized in a sparse subset of features. Building on this insight, the authors propose a fine-tuning-free latent-space intervention strategy that leverages both raw activations and latent representations from a sparse autoencoder (SAE) to detect and suppress hallucinations. Experimental results demonstrate that this approach reduces hallucination rates on non-speech test sets from 72.63% to 14.11% for Whisper-small and from 86.88% to 27.33% for Whisper-large-v3, with only a marginal increase in word error rate (WER), achieving performance comparable to fine-tuned baselines.

Abstract

Whisper, a widely adopted ASR model, is known to suffer from hallucinations - coherent transcriptions generated for non-speech audio entirely disconnected from the input. We investigate whether hallucinations can be detected and mitigated through Whisper's internal representations. We extract audio encoder activations and evaluate two representation spaces: raw Whisper activations and Sparse AutoEncoder (SAE) latents. We show that both spaces encode linearly separable hallucination-related information, with discriminative power concentrated in a sparse feature subset and increasing toward deeper encoder layers. We propose two steering strategies: activation-space steering and SAE latent-space steering. SAE-based steering reduces hallucination rate from 72.63% to 14.11% for Whisper small and from 86.88% to 27.33% for Whisper large-v3 on the full non-speech test set, with small WER degradation on speech data, approaching the performance of fine-tuning-based methods.

Problem

• Whisper
• hallucination
• ASR
• non-speech audio
• speech recognition

Innovation

• hallucination detection
• representation steering
• Sparse AutoEncoder
• Whisper ASR
• internal representation