# What Counts as an Error? Dual-Reference Benchmarking for Atypical ASR

arXiv: https://arxiv.org/abs/2606.31112 · HTML: https://mlsanity.com/paper/2606.31112/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 18 of its 25 references that are in the map.
- First posted on arXiv 30 Jun 2026.

## Builds on (most cited first, 18 of 18)

- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [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)
- [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://mlsanity.com/paper/2106.07447/) (2021)
- [Conformer: Convolution-augmented Transformer for Speech Recognition](https://mlsanity.com/paper/2005.08100/) (2020)
- [Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition](https://mlsanity.com/paper/2305.05084/) (2023)
- [QuartzNet: Deep Automatic Speech Recognition with 1D Time-Channel Separable Convolutions](https://mlsanity.com/paper/1910.10261/) (2019)
- [Canary-1B-v2 & Parakeet-TDT-0.6B-v3: Efficient and High-Performance Models for Multilingual ASR and AST](https://mlsanity.com/paper/2509.14128/) (2025)
- [Unconstrained Dysfluency Modeling for Dysfluent Speech Transcription and Detection](https://mlsanity.com/paper/2312.12810/) (2023)
- [From User Perceptions to Technical Improvement: Enabling People Who Stutter to Better Use Speech Recognition](https://mlsanity.com/paper/2302.09044/) (2023)
- [Lost in Transcription: Identifying and Quantifying the Accuracy Biases of Automatic Speech Recognition Systems Against Disfluent Speech](https://mlsanity.com/paper/2405.06150/) (2024)
- [Analysis and Tuning of a Voice Assistant System for Dysfluent Speech](https://mlsanity.com/paper/2106.11759/) (2021)
- [SeMaScore : a new evaluation metric for automatic speech recognition tasks](https://mlsanity.com/paper/2401.07506/) (2024)
- [Analysis and Evaluation of Synthetic Data Generation in Speech Dysfluency Detection](https://mlsanity.com/paper/2505.22029/) (2025)
- [Improved Robustness to Disfluencies in RNN-Transducer Based Speech Recognition](https://mlsanity.com/paper/2012.06259/) (2020)
- [Fine-Tuning ASR for Stuttered Speech: Personalized vs. Generalized Approaches](https://mlsanity.com/paper/2506.00853/) (2025)
- [Clinical Annotations for Automatic Stuttering Severity Assessment](https://mlsanity.com/paper/2506.00644/) (2025)
- [Findings of the 2024 Mandarin Stuttering Event Detection and Automatic Speech Recognition Challenge](https://mlsanity.com/paper/2409.05430/) (2024)

## Abstract (arXiv)

ASR systems have been often reported to underperform on atypical speech. An often conflated compounding factor is the existence of two valid transcription references: verbatim (actual produced speech, including repetitions/prolongations) and intended (the canonical form of the text with disfluencies removed) in atypical speech recognition depending on context and use-case. Most ASR evaluations conflate this duality into a single ground truth and reward systems that delete disfluencies, ignoring verbatim faithfulness. We benchmark 11 ASR models from encoder-decoder, CTC and transducer families using both verbatim and intended references on atypical stuttered speech as a case study. Our quantitative assessment underlines the disparity in model performance and rankings using the two transcript styles. Through this analysis, we highlight the importance of selecting a suitable transcription reference for valid model selection depending on the use-case, particularly for atypical ASR.