Careless Whisper: Speech-to-Text Hallucination Harms
Open accessThe DOI matches its Crossref record; 8 of the 8 references it cites were found in a registry or an index.
Abstract
Speech-to-text services aim to transcribe input audio as accurately as possible. They increasingly play a role in everyday life, for example in personal voice assistants or in customer-company interactions. We evaluate Open AI’s Whisper, a state-of-the-art automated speech recognition service outperforming industry competitors, as of 2023. While many of Whisper’s transcriptions were highly accurate, we find that roughly 1% of audio transcriptions contained entire hallucinated phrases or sentences which did not exist in any form in the underlying audio. We thematically analyze the Whisper-hallucinated content, finding that 38% of hallucinations include explicit harms such as perpetuating violence, making up inaccurate associations, or implying false authority. We then study why hallucinations occur by observing the disparities in hallucination rates between speakers with aphasia (who have a lowered ability to express themselves using speech and voice) and a control group. We find that hallucinations disproportionately occur for individuals who speak with longer shares of non-vocal durations—a common symptom of aphasia. We call on industry practitioners to ameliorate these language-model-based hallucinations in Whisper, and to raise awareness of potential biases amplified by hallucinations in downstream applications of speech-to-text models.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Survey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023) · ACM Computing Surveys · doi:10.1145/3571730 · Cited by 5,025
Gibbs, G. (2007) · doi:10.4135/9781849208574 · Cited by 4,299
Robust Speech Recognition via Large-Scale Weak Supervision
Radford, A., Kim, J. W., Xu, T., Brockman, G., McLeavey, C., & Sutskever, I. (2022) · arXiv · doi:10.48550/arxiv.2212.04356 · Cited by 1,953
Eysenbach, G. (2023) · JMIR Medical Education · doi:10.2196/46885 · Cited by 840 · cited 1×
“While hallucinations have been increasingly studied in the context of text generated by ChatGPT (a language model also made by OpenAI) [9, 12], hallucinations have only been considered in speech-to-text models as a means to study error prediction [31] or error identification [10], and not as a…”
Racial disparities in automated speech recognition
Koenecke, A., Nam, A., Lake, E., Nudell, J., Quartey, M., Mengesha, Z., Toups, C., Rickford, J. R., Jurafsky, D., & Goel, S. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1915768117 · Cited by 765
Self-citationAutomatic quantitative analysis of spontaneous aphasic speech
Le, D., Licata, K., & Mower Provost, E. (2018) · Speech Communication · doi:10.1016/j.specom.2018.04.001 · Cited by 97 · cited 1×
“It is critical to ensure that aphasic speakers are able to use speech-to-text tools, which tend to underperform without modeling techniques developed specifically tailored for aphasic speakers [15, 16].”
BackgroundAutomatic Speech Assessment for People with Aphasia Using TDNN-BLSTM with Multi-Task Learning
Qin, Y., Lee, T., Feng, S., & Kong, A. P. H. (2018) · Interspeech 2018 · doi:10.21437/interspeech.2018-1630 · Cited by 30 · cited 1×
“Research shows that patients exhibiting aphasia are usually older than stroke patients who do not exhibit aphasia, while those stroke patients who are younger are more likely to exhibit less fluent (e.g., non-fluent or Broca’s type) aphasia [7].”
BackgroundBringing advanced speech processing technology to the clinical management of speech disorders
McAllister, T., & Ballard, K. J. (2018) · International Journal of Speech-Language Pathology · doi:10.1080/17549507.2018.1510034 · Cited by 6
Cited by
Tang, X., Li, J., & Wu, S. (2026) · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3790405 · Cited by 4
“Meanwhile, a growing body of work has documented AI models’ degraded, and at times discriminatory, performance for users with disabilities 1 [30, 43, 52, 67, 92, 104, 115], and identified the lack of data representing disabled people as one of the root causes [12, 43, 64, 78, 88, 104].”
Looi, J. C., Allison, S., Bastiampillai, T., Reutens, S., & Looi, R. C. (2026) · Australasian Psychiatry · doi:10.1177/10398562251380544 · Cited by 4
Li, J., Li, Q., Gong, R., Wang, L., & Wu, S. (2025) · Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3715275.3732179 · Cited by 4
“…research within FAccT that has surfaced ableist biases in large language models [15] and identified technical flaws in advanced speech AI models [24], our study not only uncovers the usability and social harms faced by PWS in mainstream speech AI systems, but also showcases the promise of…”
From Tasks to Traits: Can LLMs Infer Personality From Brief Assessment Center Behavior?
Härtel, T. M. (2026) · International Journal of Selection and Assessment · doi:10.1111/ijsa.70074 · Cited by 1
Metadata
- Authors
- Koenecke, A., Choi, A. S. G., Mei, K. X., Schellmann, H., & Sloane, M.
- Venue
- The 2024 ACM Conference on Fairness, Accountability, and Transparency
- Publisher
- ACM
- Published
- 2024-06-03
- Pages
- 1672-1681
- OpenAlex
- W4391833135
- Cited by
- 89
- References
- 8
- FWCI
- 27.07
- Type
- conference paper
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Careless Whisper: Speech-to-Text Hallucination Harms. https://app.cochat.ai/papers/10.1145/3630106.3658996
- The paper
- Koenecke, A., Choi, A. S. G., Mei, K. X., Schellmann, H., & Sloane, M. (2024). Careless Whisper: Speech-to-Text Hallucination Harms. The 2024 ACM Conference on Fairness, Accountability, and Transparency. https://doi.org/10.1145/3630106.3658996
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.