pith. sign in

arxiv: 2606.11639 · v1 · pith:5PV6WMGQnew · submitted 2026-06-10 · 💻 cs.CL

Evaluating Bias in Phoneme-Based Automatic Speech Recognition Systems: An Analysis of IPA Transcription Models

Catherine Bao , Maneesha Rani Saha , Neal Patwari This is my paper

Pith reviewed 2026-06-27 10:14 UTC · model grok-4.3

classification 💻 cs.CL
keywords biasIPA transcriptionautomatic speech recognitionphoneme error ratedemographic disparitiesmultilingual ASRWhisperIPAZIPA
0
0 comments X

The pith

IPA-based ASR models show performance disparities across languages and demographic groups that persist after allowing for similar phoneme substitutions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper evaluates two state-of-the-art open-source IPA transcription systems, WhisperIPA and ZIPA, on existing multilingual speech corpora and demographically annotated English corpora. Performance is measured with standard phoneme error rate and a proposed Soft PER metric that accepts linguistically similar phoneme substitutions when comparing outputs to grapheme-to-phoneme references. The analysis finds that accuracy differences remain across languages and across groups defined by gender, accent, ethnicity, and age. A sympathetic reader would care because phoneme-based IPA layers are positioned as language-agnostic foundations for expanding ASR to low-resource languages, so any embedded disparities could propagate into many future systems. The work identifies potential sources of bias to support development of more inclusive phoneme-based recognizers.

Core claim

Our analysis examines how performance varies across languages and demographic groups such as gender, accent, ethnicity, and age, revealing persistent disparities even after accounting for acceptable phonemic variation.

What carries the argument

The Soft PER metric, which tolerates linguistically similar phoneme substitutions when scoring model IPA transcriptions against grapheme-to-phoneme reference outputs.

Load-bearing premise

The chosen grapheme-to-phoneme reference systems supply an unbiased and accurate ground truth, and the demographically annotated corpora contain representative samples without unmeasured selection effects.

What would settle it

Re-running the evaluation on the same models but with alternative G2P reference systems or on newly collected demographically balanced corpora that produces no statistically significant performance differences across the reported groups would falsify the claim of persistent disparities.

Figures

Figures reproduced from arXiv: 2606.11639 by Catherine Bao, Maneesha Rani Saha, Neal Patwari.

Figure 1
Figure 1. Figure 1: Comparison of standard PER and Soft PER across evaluated languages: (AR=Arabic, ES=Spanish, [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Effect of Soft PER on language-level evaluation for each model. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: PER and Soft PER comparison by gender across demographic datasets. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: PER and Soft PER comparison by age group [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: PER, Soft PER, and group-level disparity by ethnicity for the EdACC dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: PER, Soft PER, and group-level disparity by accent/dialect for the SVC dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

The popularization of automatic speech recognition (ASR) systems has increased exploration of the demographic biases related to race, age, gender, and accent, often formed from imbalanced training data. Most of these studies focused on standard grapheme-based ASR systems with comparatively little emphasis on phoneme-based systems, such as models that produce International Phonetic Alphabet (IPA) representations. As ASR systems shift toward multilingual support and low-resource language modeling, IPA-based layers serve as a critical, language-agnostic foundation. In this study, we evaluate the performance of two state-of-the-art open-source ASR systems, WhisperIPA and ZIPA, that generate IPA transcriptions across diverse accents and language sources. Our evaluation includes existing multilingual speech corpora and demographically annotated English-language corpora. We measure model performance by comparing model-generated IPA transcriptions against grapheme-to-phoneme (G2P) systems using both standard phoneme error rate (PER) and a proposed Soft PER metric that tolerates linguistically similar phoneme substitutions. Our analysis examines how performance varies across languages and demographic groups such as gender, accent, ethnicity, and age, revealing persistent disparities even after accounting for acceptable phonemic variation. These findings provide insight into potential sources of bias and inform the development of more inclusive and linguistically robust phoneme-based ASR systems. Our code and data will be made publicly available to the community.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The manuscript evaluates bias in two open-source phoneme-based ASR systems (WhisperIPA and ZIPA) that produce IPA transcriptions. It compares model outputs against grapheme-to-phoneme (G2P) references on multilingual speech corpora and demographically annotated English corpora, using both standard phoneme error rate (PER) and a proposed Soft PER metric that tolerates linguistically similar substitutions. The central claim is that performance varies across languages and demographic groups (gender, accent, ethnicity, age), with persistent disparities remaining even after the Soft PER adjustment for acceptable phonemic variation. The authors position the work as informing more inclusive IPA-based ASR development and commit to releasing code and data.

Significance. If the reported disparities can be shown to reflect model behavior rather than reference or metric artifacts, the study would add to fairness research in speech technology by focusing on phoneme-based, language-agnostic systems that are increasingly used for multilingual and low-resource ASR. The Soft PER proposal is a methodological step forward, and the public data release would support reproducibility. The work is timely given the shift toward IPA layers in ASR.

major comments (2)
  1. [Abstract / Evaluation] Abstract and Evaluation section: The claim of 'persistent disparities even after accounting for acceptable phonemic variation' rests on the assumption that G2P reference systems supply an unbiased ground truth. Standard G2P converters encode canonical or prestige pronunciations that systematically diverge from accented or demographically varied realizations; the Soft PER tolerance for 'linguistically similar' substitutions does not incorporate an explicit model of accent-induced variation (e.g., dialect rules or forced-alignment priors). This assumption is load-bearing for the central claim that observed PER/Soft PER gaps isolate model bias rather than reference mismatch.
  2. [Results] Results section: No sample sizes per demographic subgroup, confidence intervals, or statistical significance tests are reported for the claimed disparities. Without these quantities it is impossible to determine whether the reported differences across gender/accent/ethnicity/age are robust or could arise from sampling variation or unmeasured selection effects in the demographically annotated corpora.
minor comments (3)
  1. [Methods] The abstract lists the corpora only generically ('existing multilingual speech corpora and demographically annotated English-language corpora'); the Methods section should name the specific datasets, their sizes, and how demographic labels were obtained and validated.
  2. [Evaluation Metrics] Notation: Define 'Soft PER' formally (including the exact similarity criterion and substitution costs) in the first use rather than relying on the prose description.
  3. [Discussion] The manuscript should include a limitations paragraph addressing the possibility that G2P canonical bias affects all demographic groups unequally.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback, which identifies key areas for strengthening our analysis of bias in phoneme-based ASR. We respond to each major comment below and indicate planned revisions.

read point-by-point responses

  1. Referee: [Abstract / Evaluation] Abstract and Evaluation section: The claim of 'persistent disparities even after accounting for acceptable phonemic variation' rests on the assumption that G2P reference systems supply an unbiased ground truth. Standard G2P converters encode canonical or prestige pronunciations that systematically diverge from accented or demographically varied realizations; the Soft PER tolerance for 'linguistically similar' substitutions does not incorporate an explicit model of accent-induced variation (e.g., dialect rules or forced-alignment priors). This assumption is load-bearing for the central claim that observed PER/Soft PER gaps isolate model bias rather than reference mismatch.

    Authors: We agree that G2P systems typically reflect canonical pronunciations and that our Soft PER does not explicitly encode accent-specific rules or forced-alignment priors. This is a valid limitation on interpreting absolute bias versus reference mismatch. However, the same G2P references are applied uniformly across all demographic groups and languages, so relative performance gaps remain informative about model behavior. We will revise the Evaluation and Discussion sections to explicitly acknowledge this assumption, add caveats on the scope of our claims, and note that future work could incorporate accent-aware references. This strengthens rather than undermines the central findings. revision: partial

  2. Referee: [Results] Results section: No sample sizes per demographic subgroup, confidence intervals, or statistical significance tests are reported for the claimed disparities. Without these quantities it is impossible to determine whether the reported differences across gender/accent/ethnicity/age are robust or could arise from sampling variation or unmeasured selection effects in the demographically annotated corpora.

    Authors: We accept this point. The current manuscript omits these details, which are necessary for assessing robustness. In the revised Results section we will report subgroup sample sizes, bootstrap confidence intervals for PER and Soft PER, and appropriate statistical tests (e.g., permutation or t-tests with multiple-comparison correction) for the demographic comparisons. This will allow readers to evaluate whether observed gaps exceed sampling variation. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical evaluation

full rationale

The paper performs an empirical comparison of two IPA-based ASR systems (WhisperIPA and ZIPA) against G2P references on existing corpora, reporting PER and a defined Soft PER metric across demographic slices. No equations, derivations, fitted parameters presented as predictions, or self-citation chains appear in the abstract or described content. The central claim is an observational finding about performance disparities; it does not reduce to any input by construction. The skeptic concern addresses reference validity and metric assumptions, which is a question of external correctness rather than internal circularity in a derivation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Empirical measurement study containing no mathematical derivations, fitted constants, or postulated entities.

pith-pipeline@v0.9.1-grok · 5784 in / 974 out tokens · 25230 ms · 2026-06-27T10:14:08.696615+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

74 extracted references · 16 canonical work pages

  1. [1]

    2017 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , pages=

    Error detection of grapheme-to-phoneme conversion in text-to-speech synthesis using speech signal and lexical context , author=. 2017 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , pages=. 2017 , organization=

    2017

  2. [2]

    2024 , eprint=

    IPA Transcription of Bengali Texts , author=. 2024 , eprint=

    2024

  3. [3]

    Journal of Memory and Language , volume=

    Allophones, not phonemes in spoken-word recognition , author=. Journal of Memory and Language , volume=. 2018 , publisher=

    2018

  4. [4]

    International Journal of Language & Communication Disorders , volume=

    The agreement of phonetic transcriptions between paediatric speech and language therapists transcribing a disordered speech sample , author=. International Journal of Language & Communication Disorders , volume=. 2024 , publisher=

    2024

  5. [5]

    Linguistics Vanguard , volume=

    Assessing the accuracy of existing forced alignment software on varieties of British English , author=. Linguistics Vanguard , volume=. 2020 , publisher=

    2020

  6. [6]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Performance disparities between accents in automatic speech recognition (student abstract) , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  7. [7]

    ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Unveiling performance bias in asr systems: A study on gender, age, accent, and more , author=. ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2025 , organization=

    2025

  8. [8]

    doi:10.21437/Interspeech.2017-1746 , issn =

    Rachael Tatman and Conner Kasten , year =. doi:10.21437/Interspeech.2017-1746 , issn =

    work page doi:10.21437/interspeech.2017-1746 2017

  9. [9]

    Gender and Dialect Bias in Y ou T ube ' s Automatic Captions

    Tatman, Rachael. Gender and Dialect Bias in Y ou T ube ' s Automatic Captions. Proceedings of the First ACL Workshop on Ethics in Natural Language Processing. 2017. doi:10.18653/v1/W17-1606

    work page doi:10.18653/v1/w17-1606 2017

  10. [10]

    arXiv preprint arXiv:2502.18434 , year=

    Exploring gender disparities in automatic speech recognition technology , author=. arXiv preprint arXiv:2502.18434 , year=

    arXiv

  11. [11]

    arXiv preprint arXiv:2510.22495 , year=

    A Sociophonetic Analysis of Racial Bias in Commercial ASR Systems Using the Pacific Northwest English Corpus , author=. arXiv preprint arXiv:2510.22495 , year=

    arXiv

  12. [12]

    doi:10.21437/Interspeech.2025-1973 , issn =

    Dana Serditova and Kevin Tang and Jochen Steffens , year =. doi:10.21437/Interspeech.2025-1973 , issn =

    work page doi:10.21437/interspeech.2025-1973 2025

  13. [13]

    Linguistically Informed Tokenization Improves ASR for Underresourced Languages

    Daul, Massimo Marie and Tosolini, Alessio and Bowern, Claire. Linguistically Informed Tokenization Improves ASR for Underresourced Languages. Proceedings of the Fifth Workshop on NLP Applications to Field Linguistics. 2026. doi:10.18653/v1/2026.fieldmatters-1.4

    work page doi:10.18653/v1/2026.fieldmatters-1.4 2026

  14. [14]

    Sonos Voice Control Bias Assessment Dataset: A Methodology for Demographic Bias Assessment in Voice Assistants

    Sekkat, Chloe and Leroy, Fanny and Mdhaffar, Salima and Smith, Blake Perry and Est \`e ve, Yannick and Dureau, Joseph and Coucke, Alice. Sonos Voice Control Bias Assessment Dataset: A Methodology for Demographic Bias Assessment in Voice Assistants. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and ...

    2024

  15. [15]

    Version , volume=

    The corpus of regional african american language , author=. Version , volume=

  16. [16]

    2026 , eprint=

    Advanced Modeling of Interlanguage Speech Intelligibility Benefit with L1-L2 Multi-Task Learning Using Differentiable K-Means for Accent-Robust Discrete Token-Based ASR , author=. 2026 , eprint=

    2026

  17. [17]

    Discovering phonetic inventories with crosslingual automatic speech recognition , journal =

    Piotr Żelasko and Siyuan Feng and Laureano. Discovering phonetic inventories with crosslingual automatic speech recognition , journal =. 2022 , issn =. doi:https://doi.org/10.1016/j.csl.2022.101358 , url =

    work page doi:10.1016/j.csl.2022.101358 2022

  18. [18]

    doi:10.21437/Interspeech.2025-593 , issn =

    Kentaro Onda and Yosuke Kashiwagi and Emiru Tsunoo and Hayato Futami and Shinji Watanabe , year =. doi:10.21437/Interspeech.2025-593 , issn =

    work page doi:10.21437/interspeech.2025-593 2025

  19. [19]

    2020 , eprint=

    That Sounds Familiar: an Analysis of Phonetic Representations Transfer Across Languages , author=. 2020 , eprint=

    2020

  20. [20]

    Toward Responsible

    Cunningham, Jay L and Adjagbodjou, Adinawa and Basoah, Jeffrey and Jawara, Jainaba and Kadoma, Kowe and Lewis, Aaleyah , booktitle=. Toward Responsible

  21. [21]

    2026 , eprint=

    PRiSM: Benchmarking Phone Realization in Speech Models , author=. 2026 , eprint=

    2026

  22. [22]

    doi:10.21437/Interspeech.2023-2584 , issn =

    Chihiro Taguchi and Yusuke Sakai and Parisa Haghani and David Chiang , year =. doi:10.21437/Interspeech.2023-2584 , issn =

    work page doi:10.21437/interspeech.2023-2584 2023

  23. [23]

    International Conference on Information and Communication Technology for Intelligent Systems , pages=

    Automated International Phonetic Alphabet (IPA) Representation for Assamese Speech , author=. International Conference on Information and Communication Technology for Intelligent Systems , pages=. 2025 , organization=

    2025

  24. [24]

    Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval , pages=

    Using phoneme representations to build predictive models robust to asr errors , author=. Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval , pages=

  25. [25]

    The taste of IPA: Towards open-vocabulary keyword spotting and forced alignment in any language

    Zhu, Jian and Yang, Changbing and Samir, Farhan and Islam, Jahurul. The taste of IPA: Towards open-vocabulary keyword spotting and forced alignment in any language. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024. doi:10.18653/v1/20...

    work page doi:10.18653/v1/2024.naacl-long.43 2024

  26. [26]

    ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Leveraging IPA and articulatory features as effective inductive biases for multilingual ASR training , author=. ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2025 , organization=

    2025

  27. [27]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    ZIPA: A family of efficient models for multilingual phone recognition , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  28. [28]

    Listen, attend and spell: A neural network for large vocabulary conversational speech recognition , year=

    Chan, William and Jaitly, Navdeep and Le, Quoc and Vinyals, Oriol , booktitle=. Listen, attend and spell: A neural network for large vocabulary conversational speech recognition , year=

  29. [29]

    2020 , eprint=

    Contextual RNN-T For Open Domain ASR , author=. 2020 , eprint=

    2020

  30. [30]

    doi:10.21437/Interspeech.2022-10340 , issn =

    Fangjun Kuang and Liyong Guo and Wei Kang and Long Lin and Mingshuang Luo and Zengwei Yao and Daniel Povey , year =. doi:10.21437/Interspeech.2022-10340 , issn =

    work page doi:10.21437/interspeech.2022-10340 2022

  31. [31]

    Efficient Conformer-Based CTC Model for Intelligent Cockpit Speech Recognition , year=

    Guo, Hanzhi and Chen, Yunshu and Xie, Xukang and Xu, Gaopeng and Guo, Wei , booktitle=. Efficient Conformer-Based CTC Model for Intelligent Cockpit Speech Recognition , year=

  32. [32]

    Advances in neural information processing systems , volume=

    wav2vec 2.0: A framework for self-supervised learning of speech representations , author=. Advances in neural information processing systems , volume=

  33. [33]

    Proceedings of the 40th International Conference on Machine Learning , articleno =

    Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya , title =. Proceedings of the 40th International Conference on Machine Learning , articleno =. 2023 , publisher =

    2023

  34. [34]

    2022 , booktitle =

    Arun Babu and Changhan Wang and Andros Tjandra and Kushal Lakhotia and Qiantong Xu and Naman Goyal and Kritika Singh and Patrick. 2022 , booktitle =. doi:10.21437/Interspeech.2022-143 , issn =

    work page doi:10.21437/interspeech.2022-143 2022

  35. [35]

    and Salem, Alexandra C

    Gale, Robert C. and Salem, Alexandra C. and Fergadiotis, Gerasimos and Bedrick, Steven. Mixed Orthographic/Phonemic Language Modeling: Beyond Orthographically Restricted Transformers ( BORT ). Proceedings of the 8th Workshop on Representation Learning for NLP (RepL4NLP 2023). 2023. doi:10.18653/v1/2023.repl4nlp-1.18

    work page doi:10.18653/v1/2023.repl4nlp-1.18 2023

  36. [36]

    Evaluating Workflows for Creating Orthographic Transcripts for Oral Corpora by Transcribing from Scratch or Correcting ASR -Output

    Gorisch, Jan and Schmidt, Thomas. Evaluating Workflows for Creating Orthographic Transcripts for Oral Corpora by Transcribing from Scratch or Correcting ASR -Output. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). 2024

    2024

  37. [37]

    2020 , isbn =

    Fang, Anjie and Filice, Simone and Limsopatham, Nut and Rokhlenko, Oleg , title =. 2020 , isbn =. doi:10.1145/3397271.3401050 , booktitle =

    work page doi:10.1145/3397271.3401050 2020

  38. [38]

    Automated International Phonetic Alphabet (IPA) Representation for Assamese Speech

    Devi, Kumari Jayashree and Sarma, Parismita. Automated International Phonetic Alphabet (IPA) Representation for Assamese Speech. ICT for Intelligent Systems. 2026

    2026

  39. [39]

    arXiv preprint arXiv:2509.12647 , year=

    PAC: Pronunciation-Aware Contextualized Large Language Model-based Automatic Speech Recognition , author=. arXiv preprint arXiv:2509.12647 , year=

    arXiv

  40. [40]

    , author=

    Language identification on code-switching utterances using multiple cues. , author=. Interspeech , pages=

  41. [41]

    Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing , pages=

    Learning to predict code-switching points , author=. Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing , pages=

    2008

  42. [42]

    Oriental COCOSDA 2017 , pages=

    AIShell-1: An Open-Source Mandarin Speech Corpus and A Speech Recognition Baseline , author=. Oriental COCOSDA 2017 , pages=

    2017

  43. [43]

    2026 , eprint=

    WAXAL: A Large-Scale Multilingual African Language Speech Corpus , author=. 2026 , eprint=

    2026

  44. [44]

    arXiv preprint arXiv:2308.15710 , year=

    Speech wikimedia: A 77 language multilingual speech dataset , author=. arXiv preprint arXiv:2308.15710 , year=

    arXiv

  45. [45]

    Proceedings of the twelfth language resources and evaluation conference , pages=

    Artie bias corpus: An open dataset for detecting demographic bias in speech applications , author=. Proceedings of the twelfth language resources and evaluation conference , pages=

  46. [46]

    arXiv preprint arXiv:2408.12734 , year=

    Towards measuring fairness in speech recognition: Fair-speech dataset , author=. arXiv preprint arXiv:2408.12734 , year=

    arXiv

  47. [47]

    IPA CHILDES & G 2 P +: Feature-Rich Resources for Cross-Lingual Phonology and Phonemic Language Modeling

    Goriely, Z \'e bulon and Buttery, Paula. IPA CHILDES & G 2 P +: Feature-Rich Resources for Cross-Lingual Phonology and Phonemic Language Modeling. Proceedings of the 29th Conference on Computational Natural Language Learning. 2025. doi:10.18653/v1/2025.conll-1.33

    work page doi:10.18653/v1/2025.conll-1.33 2025

  48. [48]

    Journal of Open Source Software , volume=

    Phonemizer: Text to phones transcription for multiple languages in python , author=. Journal of Open Source Software , volume=

  49. [49]

    Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018) , year=

    Epitran: Precision G2P for many languages , author=. Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018) , year=

    2018

  50. [50]

    2025 , version =

    Taubert, Sebastian , title =. 2025 , version =. doi:10.5281/zenodo.15229718 , url =

    work page doi:10.5281/zenodo.15229718 2025

  51. [51]

    2021 , url =

    kfcd , title =. 2021 , url =

    2021

  52. [52]

    Proceedings of the Twelfth Language Resources and Evaluation Conference , pages=

    AlloVera: A multilingual allophone database , author=. Proceedings of the Twelfth Language Resources and Evaluation Conference , pages=

  53. [53]

    , author=

    Montreal forced aligner: Trainable text-speech alignment using kaldi. , author=. Interspeech , volume=

  54. [54]

    Journal of Phonetics , volume=

    Introducing parselmouth: A python interface to praat , author=. Journal of Phonetics , volume=. 2018 , publisher=

    2018

  55. [55]

    The Edinburgh International Accents of English Corpus: Towards the Democratization of English ASR , year=

    Sanabria, Ramon and Bogoychev, Nikolay and Markl, Nina and Carmantini, Andrea and Klejch, Ondrej and Bell, Peter , booktitle=. The Edinburgh International Accents of English Corpus: Towards the Democratization of English ASR , year=

  56. [56]

    arXiv preprint arXiv:2503.20212 , year=

    Dolphin: A large-scale automatic speech recognition model for eastern languages , author=. arXiv preprint arXiv:2503.20212 , year=

    arXiv

  57. [57]

    2024 , eprint=

    Anatomy of Industrial Scale Multilingual ASR , author=. 2024 , eprint=

    2024

  58. [58]

    arXiv preprint arXiv:2511.09690 , year=

    Omnilingual ASR: Open-Source Multilingual Speech Recognition for 1600+ Languages , author=. arXiv preprint arXiv:2511.09690 , year=

    arXiv

  59. [59]

    OWSM - CTC : An Open Encoder-Only Speech Foundation Model for Speech Recognition, Translation, and Language Identification

    Peng, Yifan and Sudo, Yui and Shakeel, Muhammad and Watanabe, Shinji. OWSM - CTC : An Open Encoder-Only Speech Foundation Model for Speech Recognition, Translation, and Language Identification. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024. doi:10.18653/v1/2024.acl-long.549

    work page doi:10.18653/v1/2024.acl-long.549 2024

  60. [60]

    2023 , publisher =

    ctaguchi , title =. 2023 , publisher =

    2023

  61. [61]

    2025 , eprint=

    POWSM: A Phonetic Open Whisper-Style Speech Foundation Model , author=. 2025 , eprint=

    2025

  62. [62]

    2020 , publisher =

    Facebook AI , title =. 2020 , publisher =

    2020

  63. [63]

    2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , pages=

    Exploring data augmentation in bias mitigation against non-native-accented speech , author=. 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , pages=. 2023 , organization=

    2023

  64. [64]

    Speech Communication , volume=

    Uneven success: automatic speech recognition and ethnicity-related dialects , author=. Speech Communication , volume=. 2022 , publisher=

    2022

  65. [65]

    , author=

    Understanding Racial Disparities in Automatic Speech Recognition: The Case of Habitual" be". , author=. Interspeech , pages=

  66. [66]

    Proceedings of the national academy of sciences , volume=

    Racial disparities in automated speech recognition , author=. Proceedings of the national academy of sciences , volume=. 2020 , publisher=

    2020

  67. [67]

    Proceedings of the 12th Workshop on NLP for Similar Languages, Varieties and Dialects , pages=

    Adapting Whisper for regional dialects: Enhancing public services for vulnerable populations in the United Kingdom , author=. Proceedings of the 12th Workshop on NLP for Similar Languages, Varieties and Dialects , pages=

  68. [68]

    doi:10.21437/Interspeech.2022-836 , issn =

    Yuanyuan Zhang and Yixuan Zhang and Bence Halpern and Tanvina Patel and Odette Scharenborg , year =. doi:10.21437/Interspeech.2022-836 , issn =

    work page doi:10.21437/interspeech.2022-836 2022

  69. [69]

    Applied Sciences , volume=

    Automatic speech recognition (asr) systems for children: A systematic literature review , author=. Applied Sciences , volume=. 2022 , publisher=

    2022

  70. [70]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Hey, siri! why are you biased against women?(student abstract) , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  71. [71]

    A literature review , author=

    Hey ASR system! Why aren’t you more inclusive? Automatic speech recognition systems’ bias and proposed bias mitigation techniques. A literature review , author=. International conference on human-computer interaction , pages=. 2022 , organization=

    2022

  72. [72]

    arXiv preprint arXiv:2103.15122 , year=

    Quantifying bias in automatic speech recognition , author=. arXiv preprint arXiv:2103.15122 , year=

    arXiv

  73. [73]

    International Conference on Learning Representations , volume=

    Zipformer: A faster and better encoder for automatic speech recognition , author=. International Conference on Learning Representations , volume=

  74. [74]

    2021 , eprint=

    MediaSpeech: Multilanguage ASR Benchmark and Dataset , author=. 2021 , eprint=

    2021