# Learning Noise-Invariant Representations for Robust Speech Recognition

arXiv: https://arxiv.org/abs/1807.06610 · HTML: https://mlsanity.com/paper/1807.06610/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 17 Jul 2018.

## Built on by (most cited first, 10 of 10)

- [Recent Advances in End-to-End Automatic Speech Recognition](https://mlsanity.com/paper/2111.01690/) (2021)
- [SPIRAL: Self-supervised Perturbation-Invariant Representation Learning for Speech Pre-Training](https://mlsanity.com/paper/2201.10207/) (2022)
- [Robust learning from corrupted EEG with dynamic spatial filtering](https://mlsanity.com/paper/2105.12916/) (2021)
- [Toward Learning Robust and Invariant Representations with Alignment Regularization and Data Augmentation](https://mlsanity.com/paper/2206.01909/) (2022)
- [An Investigation of End-to-End Models for Robust Speech Recognition](https://mlsanity.com/paper/2102.06237/) (2021)
- [On-the-Fly Aligned Data Augmentation for Sequence-to-Sequence ASR](https://mlsanity.com/paper/2104.01393/) (2021)
- [Cross-domain Speech Recognition with Unsupervised Character-level Distribution Matching](https://mlsanity.com/paper/2104.07491/) (2021)
- [Scenario Aware Speech Recognition: Advancements for Apollo Fearless Steps & CHiME-4 Corpora](https://mlsanity.com/paper/2109.11086/) (2021)
- [Robust Channel Learning for Large-Scale Radio Speaker Verification](https://mlsanity.com/paper/2406.10956/) (2024)
- [Deepfake Detection of Singing Voices With Whisper Encodings](https://mlsanity.com/paper/2501.18919/) (2025)

## Abstract (arXiv)

Despite rapid advances in speech recognition, current models remain brittle to superficial perturbations to their inputs. Small amounts of noise can destroy the performance of an otherwise state-of-the-art model. To harden models against background noise, practitioners often perform data augmentation, adding artificially-noised examples to the training set, carrying over the original label. In this paper, we hypothesize that a clean example and its superficially perturbed counterparts shouldn't merely map to the same class --- they should map to the same representation. We propose invariant-representation-learning (IRL): At each training iteration, for each training example,we sample a noisy counterpart. We then apply a penalty term to coerce matched representations at each layer (above some chosen layer). Our key results, demonstrated on the Librispeech dataset are the following: (i) IRL significantly reduces character error rates (CER) on both 'clean' (3.3% vs 6.5%) and 'other' (11.0% vs 18.1%) test sets; (ii) on several out-of-domain noise settings (different from those seen during training), IRL's benefits are even more pronounced. Careful ablations confirm that our results are not simply due to shrinking activations at the chosen layers.