💬 NLP

TMASC: Transmasculine Attitude and Speech Corpus

This paper introduces the Transmasculine Attitudes and Speech Corpus (TMASC), a multimodal dataset comprising questionnaire responses and audio recordings from 196 transmasculine individuals, and demonstrates its utility through case studies on vocal health, group characteristics, and acoustic measurement calibration.

Original authors: Sidney Wong

Published 2026-06-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Sidney Wong

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have a choir of 196 singers, but instead of just listening to them sing, you also get to ask them how they feel about their voices, what their daily lives are like, and what they wish their voices sounded like. This paper introduces a new "voice library" called TMASC (Transmasculine Attitudes and Speech Corpus). It is a collection of data specifically designed to understand the voices of transmasculine people (people assigned female at birth who identify as men or trans-masculine).

Here is a breakdown of what the paper does, using simple analogies:

1. The Problem: The "One-Size-Fits-All" Map

For a long time, doctors and researchers treated the voice needs of trans people like a map with only one path. They assumed that if a transmasculine person took testosterone (a hormone therapy), their voice would automatically drop low enough to sound "masculine," and that would be the end of the story.

The paper argues this is like assuming that because everyone has a car, they all drive the same way. In reality, transmasculine people use many different "vehicles" to get where they want to go: hormone therapy, chest binding (wrapping the chest to flatten it), surgery, or even habits like smoking to lower their pitch. The paper says we need a better map that accounts for all these different routes, not just the hormone one.

2. The Solution: A Crowd-Sourced "Voice Diary"

To build this better map, the researchers created a digital "voice diary" using a crowd-sourcing approach. Instead of asking people to come into a sterile, quiet laboratory (which can be expensive and intimidating), they asked people to record themselves at home using their own laptops or phones.

The Participants: 196 people filled out a detailed questionnaire.
The Audio: 66 of those people also recorded their voices. They didn't just read a story; they also coughed and cleared their throats (to check vocal health) and read a famous fable ("The North Wind and the Sun") in English or German.
The Questionnaire: This was a 60-question survey asking about their life, their gender identity, how satisfied they were with their voice, and whether they used testosterone or other methods to change their voice.

3. What They Found: Three "Case Studies"

The paper doesn't just list numbers; it shows three ways this data can be used, like looking at a photo album in three different ways:

Case Study 1: The "Feeling" vs. The "Fact"
The researchers looked at how people felt about their voices versus what the audio actually sounded like. They found that a person's feeling of "masculinity" in their voice didn't always match the exact pitch (how high or low the note is) on a graph. It's like how a person might feel very "tall" even if they are average height; the internal feeling and the external measurement don't always line up perfectly.
Case Study 2: Setting the "Community Standard"
Usually, doctors compare a patient's voice to "average cisgender men" (men who were born male). This paper suggests that's like comparing a unique, custom-built house to a standard cookie-cutter house. Instead, TMASC helps establish what a "normal" voice looks like within the transmasculine community itself. They found that people who were very satisfied with their voices tended to have lower pitches, but the relationship wasn't a straight line—it was more complex.
Case Study 3: Checking the "Ruler"
The researchers used the data to test two different computer programs (Praat and REAPER) that measure voice pitch. They found that the two programs gave different numbers for the same voice recordings. It's like using two different rulers where one says a table is 5 feet long and the other says 5.5 feet. This is important because if researchers use the wrong "ruler," they might get the wrong idea about what a healthy voice sounds like.

4. The Catch: It's Not Perfect

The paper is honest about the limitations, comparing the data to a snapshot taken on a busy street rather than a controlled studio photo:

Not a Time-Lapse: This is a single snapshot in time. It doesn't follow people over years to see how their voices change day-by-day.
Background Noise: Because people recorded at home, there might be background noise (like a dog barking or traffic) that isn't present in a lab.
Language Barrier: Most of the data came from English and German speakers, even though the survey was available in many languages.

The Bottom Line

The paper isn't trying to invent a new medical test for doctors to use immediately. Instead, it's building a community resource. It's like gathering a massive collection of stories and voice samples to help researchers and community members understand that there is no single "right" way for a transmasculine voice to sound. It provides a new, more accurate way to measure and understand these voices, moving away from strict medical rules and toward community-based realities.

Technical Summary: Transmasculine Attitudes and Speech Corpus (TMASC)

Problem Statement
Research into the vocal needs of transmasculine individuals has historically received less attention than that of transfeminine individuals. This disparity stems from a prevailing assumption that testosterone replacement therapy (TRT) sufficiently lowers fundamental frequency ( $f_0$ ) to achieve a gender-congruent voice. However, vocal health in transmasculine populations is influenced by a complex interplay of factors beyond TRT, including chest binding, top surgery, glottoplasty, and informal practices like smoking. Furthermore, existing diagnostic tools (e.g., VHI, TSEQ) were largely developed for transfeminine needs, and current clinical benchmarks often rely on cisgender population norms rather than community-specific standards. There is a lack of high-quality, multimodal data that integrates psychosocial self-perception with acoustic measurements to understand the vocal needs of transmasculine individuals at a community level.

Methodology
The authors developed the Transmasculine Attitudes and Speech Corpus (TMASC), a crowd-sourced multimodal dataset collected between July and October 2017.

Participants: The study included 196 transmasculine individuals (aged 22–37.5 on average) recruited via online email and social media networks. Participants were primarily from English-speaking countries (USA, Australia, NZ, Canada, UK) and German-speaking countries (Germany, Switzerland).
Data Collection: Data was gathered using the Language, Brain and Behaviour Corpus Analysis Tool (LaBB-CAT) via participants' personal electronic devices.
- Questionnaire: A 60-item survey addressed self-perception, vocal satisfaction, communicative factors, TRT usage history, vocal intervention methods, and demographics.
- Speech Samples: 66 participants provided optional audio recordings, including throat-clearing, coughing, and a reading of "The North Wind and the Sun" in English or German.
Analysis: The study utilized R for visualization and analysis. Acoustic measurements (mean and mode $f_0$ ) were extracted using both Praat (via LaBB-CAT) and the Robust Epoch And Pitch EstimatoR (REAPER) to compare pitch-tracking algorithms.

Key Contributions

Dataset Creation: The introduction of TMASC, a unique resource containing 196 questionnaire responses and 66 multimodal speech samples, specifically targeting the under-researched transmasculine population.
Multimodal Integration: The corpus links self-reported psychosocial data (vocal satisfaction, perceived masculinity) with acoustic data, allowing for the investigation of socio-acoustic norms outside clinical settings.
Methodological Validation: The paper demonstrates the feasibility of using crowd-sourced, home-recorded data to establish community-level benchmarks, offering a cost-effective alternative to laboratory-based data collection.

Results and Case Studies
The paper presents three case studies to illustrate the utility of the corpus:

Self-Perceived Vocal Health: Analysis of the 196 questionnaires revealed a holistic view of vocal satisfaction. While many participants perceived their voices as "in-between," there was a noted interaction between self-perceived vocal masculinity and personal impacts (e.g., feeling authentic or frustrated).
Community-Level Benchmarks: The study found a weak linear relationship between self-perceived vocal masculinity and mean $f_0$ . However, a more linear relationship was observed between vocal satisfaction and mean $f_0$ . Density plots showed that participants satisfied with their voices exhibited a unimodal distribution of mean $f_0$ (peak below 100 Hz), whereas those "somewhat satisfied" showed a bimodal distribution (peaks around 100 Hz and 130 Hz).
Calibration of Acoustic Measures: A comparison of pitch-tracking tools revealed significant systematic variation. REAPER yielded significantly lower mean $f_0$ values (119.1 Hz) compared to Praat (150.7 Hz) for the same samples. This highlights the necessity of calibrating acoustic measures based on the specific algorithms used when establishing community norms.

Significance and Limitations
The paper claims that TMASC provides a valuable source of perceptual and acoustic information to understand transmasculine vocal needs, moving beyond the limitations of clinical diagnostic tools. It argues that crowd-sourcing is a viable method for generating community-appropriate benchmarks that might otherwise be missed in clinical environments.

The authors explicitly acknowledge several limitations:

Cross-Sectional Design: The study is not longitudinal and cannot track individual changes over time or the specific individual effects of TRT.
Environmental Noise: Data collected in non-laboratory settings may be affected by background noise or faulty equipment, though the authors note that participants may feel more at ease in familiar environments.
Demographic Constraints: Despite global recruitment, the speech sample was heavily skewed toward English-speaking participants, and the English-only questionnaire may have excluded non-English speakers.

The authors conclude that while the study has limitations, it suggests a need for systemic changes in managing transmasculine vocal health and provides a foundation for future research into population trends.