Audio preprocessing reduces stuttering detection accuracy especially for blocks
Audio Preprocessing Effects on Stuttering Detection: A Class-Specific Analysis
Sound
Summary
Detecting stuttering in speech can be harder when the audio is processed through common cleaning and compression steps. The authors tested how these steps affect a system’s ability to spot different types of stuttering using a speech dataset. They found that certain stutter types, like long blocks of sound, are harder to detect after processing. Adjusting detection settings helps regain some accuracy, but not all measures improve even with retraining. This means some audio cleanup methods can make stuttering detection less reliable.
What this means in practice
- •For speech technology developers: Improve stuttering detection tools by adjusting detection thresholds after audio preprocessing to maintain accuracy in real-world conditions.
- •For audio codec engineers: Design audio compression methods that minimize negative impact on speech disorder detection systems by understanding which processing steps degrade detection most.
Tested on one dataset.
Authors
Anisha Pattanayak, Hanie Kang, Huang-Cheng Chou, Sudarsana Reddy Kadiri
Abstract
Audio preprocessing can affect how well a system detects stuttering. We study a simulated chain of denoising,loudness normalisation, Opus coding, and voice activity detection on SEP-28k. We use frozen WavLM Base+ features and report pointwise confidence intervals from episode-level bootstrap resampling. At a fixed threshold of 0.5, the chain reduces block F1 from 0.638 to 0.465, with smaller decreases for the other four classes. ROC-AUC decreases for all five classes. Blocks show the largest F1 and ROC-AUC losses, while sound repetitions show the largest average precision loss. Tuning the threshold on processed validation audio raises block F1 to 0.630. Retraining on processed audio with threshold tuning gives 0.628. Threshold adjustment therefore accounts for most of the observed block F1 recovery. It does not change ROC-AUC, which retraining raises only from 0.620 to 0.633, compared with 0.724 on clean audio.