Towards Microphone-Invariant Speech Representations for Transformer-Based Voice Disorder Detection
Organizations: Cognitive Systems Lab, University of Bremen, Germany
Abstract
Automatic voice disorder detection from speech offers strong potential for early diagnosis and continuous disease monitoring. While models based on pre-trained transformers have improved detection performance, their robustness to microphone variation remains underexplored, particularly for consumer-grade devices. To the best of our knowledge, we present the first study of microphone mismatch in transformer-based binary voice disorder detection across 20 smartphone microphones. While most mismatched microphone conditions cause moderate performance degradation, the worst-case condition reduces unweighted average recall by up to 12.53 percentage points. To address this, we introduce a Barlow Twins-inspired module to learn microphone-invariant representations. Our approach reduces the performance gap between matched and mismatched conditions by up to 62%, while maintaining comparable matched-condition performance. Qualitative analysis of the embedding space demonstrates more compact, distinct speaker clusters across microphones. Importantly, results indicate generalisation to unseen microphones across manufacturers, smartphone generations, and release periods, without requiring any inference-time adaptation.
Figures & tables
| Microphone condition | ||||
| Train device subset | Base | Worst | All | |
| Wav2Vec | OnePerMfr (10/20) | 0.16$$ | 7.07$$ | 1.81 ( $-$55 % std.) |
| FirstGen (5/20) | 0.56$$ | 8.52$$ | 1.18 ( $-$71 % std.) | |
| Pre2022 (17/20) | 0.08$$ | 8.26$$ | 1.56 ( $-$65 % std.) | |
| AllButOne (19/20) | 0.40$$ | 7.88$$ | 1.23 ( $-$68 % std.) | |
| OneMfrOnly (3/20) | 0.57$$ | 4.82$$ | 2.01 ( $-$34 % std.) | |
Speech Processing
Speech processing develops computational methods for recognizing, synthesizing, analyzing, and transforming spoken audio.
See all 422 papers