# Voice, Bias, and Coreference: An Interpretability Study of Gender in Speech Translation

arXiv: https://arxiv.org/abs/2511.21517 · HTML: https://mlsanity.com/paper/2511.21517/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 54 of its 90 references that are in the map.
- First posted on arXiv 26 Nov 2025.

## Builds on (most cited first, 30 of 54)

- [Language (Technology) is Power: A Critical Survey of "Bias" in NLP](https://mlsanity.com/paper/2005.14050/) (2020)
- [AudioPaLM: A Large Language Model That Can Speak and Listen](https://mlsanity.com/paper/2306.12925/) (2023)
- [Evaluating Gender Bias in Machine Translation](https://mlsanity.com/paper/1906.00591/) (2019)
- [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://mlsanity.com/paper/2010.05171/) (2020)
- [On decoder-only architecture for speech-to-text and large language model integration](https://mlsanity.com/paper/2307.03917/) (2023)
- [Assessing Gender Bias in Machine Translation -- A Case Study with Google Translate](https://mlsanity.com/paper/1809.02208/) (2018)
- [Hybrid Autoregressive Transducer (hat)](https://mlsanity.com/paper/2003.07705/) (2020)
- [Getting Gender Right in Neural Machine Translation](https://mlsanity.com/paper/1909.05088/) (2019)
- [Context-Aware Neural Machine Translation Learns Anaphora Resolution](https://mlsanity.com/paper/1805.10163/) (2018)
- [Reducing Gender Bias in Neural Machine Translation as a Domain Adaptation Problem](https://mlsanity.com/paper/2004.04498/) (2020)
- [Internal Language Model Estimation for Domain-Adaptive End-to-End Speech Recognition](https://mlsanity.com/paper/2011.01991/) (2020)
- [Inseq: An Interpretability Toolkit for Sequence Generation Models](https://mlsanity.com/paper/2302.13942/) (2023)
- [Equalizing Gender Biases in Neural Machine Translation with Word Embeddings Techniques](https://mlsanity.com/paper/1901.03116/) (2019)
- [Gender in Danger? Evaluating Speech Translation Technology on the MuST-SHE Corpus](https://mlsanity.com/paper/2006.05754/) (2020)
- [Quantifying Bias in Automatic Speech Recognition](https://mlsanity.com/paper/2103.15122/) (2021)
- [Lost in Translation: Loss and Decay of Linguistic Richness in Machine Translation](https://mlsanity.com/paper/1906.12068/) (2019)
- [Towards Holistic Evaluation of Large Audio-Language Models: A Comprehensive Survey](https://mlsanity.com/paper/2505.15957/) (2025)
- [Probing phoneme, language and speaker information in unsupervised speech representations](https://mlsanity.com/paper/2203.16193/) (2022)
- [First the worst: Finding better gender translations during beam search](https://mlsanity.com/paper/2104.07429/) (2021)
- [Towards Measuring Fairness in Speech Recognition: Casual Conversations Dataset Transcriptions](https://mlsanity.com/paper/2111.09983/) (2021)
- [Investigating Methods to Improve Language Model Integration for Attention-based Encoder-Decoder ASR Models](https://mlsanity.com/paper/2104.05544/) (2021)
- [Twists, Humps, and Pebbles: Multilingual Speech Recognition Models Exhibit Gender Performance Gaps](https://mlsanity.com/paper/2402.17954/) (2024)
- [Gender Representation in French Broadcast Corpora and Its Impact on ASR Performance](https://mlsanity.com/paper/1908.08717/) (2019)
- [Under the Morphosyntactic Lens: A Multifaceted Evaluation of Gender Bias in Speech Translation](https://mlsanity.com/paper/2203.09866/) (2022)
- [Analysing Neural Language Models: Contextual Decomposition Reveals Default Reasoning in Number and Gender Assignment](https://mlsanity.com/paper/1909.08975/) (2019)
- [Don't speak too fast: The impact of data bias on self-supervised speech models](https://mlsanity.com/paper/2110.07957/) (2021)
- [Hi Guys or Hi Folks? Benchmarking Gender-Neutral Machine Translation with the GeNTE Corpus](https://mlsanity.com/paper/2310.05294/) (2023)
- [Evaluating Gender Bias in Speech Translation](https://mlsanity.com/paper/2010.14465/) (2020)
- [A Tale of Pronouns: Interpretability Informs Gender Bias Mitigation for Fairer Instruction-Tuned Machine Translation](https://mlsanity.com/paper/2310.12127/) (2023)
- [Gender Aware Spoken Language Translation Applied to English-Arabic](https://mlsanity.com/paper/1802.09287/) (2018)

## Built on by (most cited first, 5 of 5)

- [What Triggers my Model? Contrastive Explanations Inform Gender Choices by Translation Models](https://mlsanity.com/paper/2512.08440/) (2025)
- [Why We Need Speech to Evaluate Speech Translation](https://mlsanity.com/paper/2605.28227/) (2026)
- [Voice or Stereotype? Disentangling Acoustic and Content-Based Gender in Speech-to-Speech Models](https://mlsanity.com/paper/2609.09263/) (2026)
- [Explaining GAND: A Resource on Gender-Ambiguous Natural Data & Contrastive Attribution](https://mlsanity.com/paper/2607.22546/) (2026)
- [Symphony of Bias: Exploring Gender Associations with Musical Instruments in Multimodal LLMs](https://mlsanity.com/paper/2607.26355/) (2026)

## Abstract (arXiv)

Unlike text, speech conveys information about the speaker, such as gender, through acoustic cues like pitch. This gives rise to modality-specific bias concerns. For example, in speech translation (ST), when translating from languages with notional gender, such as English, into languages where gender-ambiguous terms referring to the speaker are assigned grammatical gender, the speaker's vocal characteristics may play a role in gender assignment. This risks misgendering speakers, whether through masculine defaults or vocal-based assumptions. Yet, how ST models make these decisions remains poorly understood. We investigate the mechanisms ST models use to assign gender to speaker-referring terms across three language pairs (en-es/fr/it). To do so, we examine how training data patterns, internal language model (ILM) biases, and acoustic information interact. We find that models do not simply replicate term-specific gender associations from training data, but learn broader patterns of masculine prevalence. While the ILM exhibits strong masculine bias, models can override these preferences based on acoustic input. Using contrastive feature attribution on spectrograms, we reveal that the model with higher gender accuracy relies on a previously unknown mechanism: using first-person pronouns to link gendered terms back to the speaker, accessing gender information distributed across the frequency spectrum rather than concentrated in pitch.