# Perceptually Better, Semantically Worse: Measuring Speech Enhancement Impact on LLM-Based Voice Systems

arXiv: https://arxiv.org/abs/2608.30348 · HTML: https://mlsanity.com/paper/2608.30348/

## Key facts

- No citing papers in the map yet (data through 19 Oct 2026).
- Placed in Speech enhancement and noise suppression, one of 11 directions in Speech and audio language models.
- Builds on 12 of its 19 references that are in the map.
- First posted on arXiv 31 Aug 2026.

## Placed in

- Speech and audio language models / [Speech enhancement and noise suppression](https://mlsanity.com/speech/speech-enhancement-and-noise-suppression/)

## Builds on (most cited first, 12 of 12)

- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://mlsanity.com/paper/2006.11477/) (2020)
- [VoiceBench: Benchmarking LLM-Based Voice Assistants](https://mlsanity.com/paper/2410.17196/) (2024)
- [SLURP: A Spoken Language Understanding Resource Package](https://mlsanity.com/paper/2011.13205/) (2020)
- [MetricGAN+: An Improved Version of MetricGAN for Speech Enhancement](https://mlsanity.com/paper/2104.03538/) (2021)
- [Interspeech 2021 Deep Noise Suppression Challenge](https://mlsanity.com/paper/2101.01902/) (2021)
- [URGENT Challenge: Universality, Robustness, and Generalizability For Speech Enhancement](https://mlsanity.com/paper/2406.04660/) (2024)
- [Interspeech 2025 URGENT Speech Enhancement Challenge](https://mlsanity.com/paper/2505.23212/) (2025)
- [Acoustic echo cancellation with the dual-signal transformation LSTM network](https://mlsanity.com/paper/2010.14337/) (2020)
- [DENOASR: Debiasing ASRs through Selective Denoising](https://mlsanity.com/paper/2410.16712/) (2024)
- [D4AM: A General Denoising Framework for Downstream Acoustic Models](https://mlsanity.com/paper/2311.16595/) (2023)
- [When De-noising Hurts: A Systematic Study of Speech Enhancement Effects on Modern Medical ASR Systems](https://mlsanity.com/paper/2512.17562/) (2025)

## Abstract (arXiv)

Speech enhancement (SE) is commonly applied as a preprocessing step in spoken AI pipelines under the assumption that better audio quality improves downstream task performance. Whether SE-induced distortions propagate to downstream LLM task performance remains an open question. We introduce Output Divergence Rate (ODR), which measures how often SE changes an LLM's intent classification relative to clean speech, and benchmark five conditions on 2,974 SLURP clips using Whisper large-v3 and wav2vec2-large cascades. Every condition produces ODR significantly above zero (p < 0.001, binomial test). MetricGAN+ more than doubles ODR versus unenhanced noisy speech (0.318 vs. 0.135) despite improving PESQ, and unmitigated echo reaches an ODR of 0.836 through speaker substitution, a failure WER cannot capture. Audio quality metrics range from near-zero to moderate correlation with ODR (SQUIM-MOS ρ=-0.068, PESQ ρ=-0.467). The MetricGAN+ and echo results replicate across ASR architectures, indicating that standard audio quality metrics are insufficient for LLM pipeline quality.