EMNLP Findings2025
Fairness in Automatic Speech Recognition Isn't a One-Size-Fits-All
Hend Elghazaly, Bahman Mirheidari, Heidi Christensen, Nafise Sadat Moosavi
摘要
Modern Automatic Speech Recognition (ASR) systems are increasingly deployed in highstakes settings, including clinical interviews, public services, and educational tools, where equitable performance across speaker groups is essential. While pre-trained speech models like Whisper achieve strong overall accuracy, they often exhibit inconsistent grouplevel performance that varies across domains. These disparities are not fixed properties of the model, but emerge from the interaction between model, data, and task-posing challenges for fairness interventions designed indomain. We frame fairness in ASR as a generalisation problem. We fine-tune a Whisper model on the Fair-Speech corpus using four strategies: basic fine-tuning, demographic rebalancing, gender-swapped data augmentation, and a novel contrastive learning objective that encourages gender-invariant representations. We evaluate performance across multiple aspects of fairness and utility, both in-domain and on three out-of-domain test sets: LibriSpeech, EdAcc, and CognoSpeak. Our findings show that the method with the best in-domain fairness performed worst out-of-domain, illustrating that fairness gains do not always generalise. Demographic balancing generalises more consistently, while our contrastive method offers a practical alternative: it achieves stable, crossdomain fairness improvements without requiring changes to the training data distribution, and with minimal accuracy trade-offs.
问 Lune
几分钟读透这篇 Paper,不用再花一下午。
Lune 能读完全文,也能读引用它的顶会 Paper,讲清它的结论、证据和局限,每一点都注明出自哪篇 Paper。不用硬啃,也不靠猜。
脉络
- Robust Speech Recognition via Large-Scale Weak Supervision 引用了 wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations
- FairFil: Contrastive Neural Debiasing Method for Pretrained Text Encoders 引用了 A Simple Framework for Contrastive Learning of Visual Representations
- wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations 引用了 A Simple Framework for Contrastive Learning of Visual Representations
- Understanding the Behaviour of Contrastive Loss 引用了 A Simple Framework for Contrastive Learning of Visual Representations
它引用的 Paper5
- A Simple Framework for Contrastive Learning of Visual RepresentationsTing Chen, Simon Kornblith, Mohammad Norouzi, Geoffrey E. HintonICML 2020 · 24,064 次引用
- wav2vec 2.0: A Framework for Self-Supervised Learning of Speech RepresentationsAlexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, Michael AuliNeurIPS 2020 · 9,451 次引用
- Robust Speech Recognition via Large-Scale Weak SupervisionAlec Radford, Jong Wook Kim, Tao Xu, Greg Brockman 等ICML 2023 · 6,966 次引用
- FairFil: Contrastive Neural Debiasing Method for Pretrained Text EncodersPengyu Cheng, Weituo Hao, Siyang Yuan, Shijing Si 等ICLR 2021 · 50 次引用
- Understanding the Behaviour of Contrastive LossFeng Wang, Huaping LiuCVPR 2021
相关 Paper
- Identifying and typifying demographic unfairness in phoneme-level embeddings of self-supervised speech recognition modelsFelix Herron, Solange Rossato, Alexandre Allauzen, François PortetACL Findings 2026
- "It feels like we're not meeting the criteria": Examining and Mitigating the Cascading Effects of Bias in Automatic Speech Recognition in Spoken Language InterfacesKelechi Ezema, Chelsea Chandler, Rosy Southwell, Niranjan Cholendiran 等CHI 2025 · 8 次引用
- Learning Fair Representation via Distributional Contrastive DisentanglementChangdae Oh, Heeji Won, Junhyuk So, Taero Kim 等KDD 2022 · 29 次引用
- Pseudo2Real: Task Arithmetic for Pseudo-Label Correction in Automatic Speech RecognitionYi-Cheng Lin, Yu-Hsuan Li Liang, Hsuan Su, Tzu-Quan Lin 等ACL Findings 2026
- Fairness Beyond Performance: Revealing Reliability Disparities Across Groups in Legal NLPT. Y. S. S. Santosh, Irtiza ChowdhuryACL 2025
