Is your smartphone biased against women? The emergence of voice assistant technology—and long-standing disparities regarding gender in the tech sector—has led to increased scrutiny of assistants like Siri, Alexa, and Cortana, with accusations that they reinforce gender stereotypes and prompt users to view women as subordinate. But does this technology actually promote gender discrimination, or does it simply mirror existing biases?
Concerns about gendered voice assistants emerged alongside the highly publicized releases of proprietary voices—many of which are female—linked to smartphones and other devices. The backlash has intensified as the adoption of voice assistants is expected to reach 8.4 billion users globally by the end of 2024.
Why are there concerns? “Obedient and obliging machines that imitate women are entering our homes, vehicles, and workplaces,” stated Saniye Gülser Corat, director of gender equality at UNESCO, in a statement accompanying a 2019 UNESCO report on voice assistants and gender.
“Their inherent subservience shapes how individuals interact with female voices and models how women reply to requests and communicate.”
The UN agency indicates that “with few exceptions,” voice assistants are predominantly designed to be female—an intentional choice by tech companies citing market research showing a preference for female voices.
Immediate reactions
“Voice is incredibly complex—it has many dimensions,” explains Naim Zirau, who co-authored a 2021 study regarding voice assistants, gender, and pitch, while serving as a researcher at the University of St. Gallen in Switzerland. Elements such as age, the gender of the listener, demographic variables, and their perception of the voice assistant’s function all seem to influence our responses to gendered voices.
Zirau’s research team created a voice interface and assigned participants the tasks of booking a flight or completing a financial survey with voices of different pitches—typically linked to gender as male voices are often lower in pitch due to hormonal influences. Their findings revealed that participants made quick judgments about the gender of a digital assistant when they heard each voice—and simply hearing a voice associated with a specific gender led them to form stereotypical, gendered assumptions about the assistant.
Listeners were more inclined to attribute traits like “delicate” and “empathetic” to female-voiced assistants, while those with male-voiced assistants were perceived as “dominant.” In addition, participants consistently assigned a gender to a gender-neutral voice, even when offered the choice to indicate uncertainty. “People have a very dichotomous understanding of gender,” notes Zirau.
Some studies suggest we can identify the gender of a speaker in as little as five seconds, yet our preferences can often be contradictory. For instance, a 2020 survey found that while a majority of both genders preferred hearing a female voice on their smart speakers, regional accents significantly influenced perceptions of credibility, regardless of the speaker’s gender. Additional research indicates that people’s trust in different gendered voices changes depending on the social situation.
The “women-are-wonderful” phenomenon
What drives the tendency to categorize a computerized voice as male or female? Chris Mayhorn, head of the psychology department at North Carolina State University, attributes this to social norms—and a longstanding instinct to personify machines. “When individuals hear a voice, they almost instinctively apply social norms,” he states, which includes a binary understanding of gender.
Mayhorn collaborated on a recent study that explored how perceived gender influences participants’ views on voice assistants. Generally, participants exhibited more trust in a female voice for medical advice and tended to view female voices as “more benevolent” compared to male voices—this social bias is sometimes referred to as the “women-are-wonderful effect.”
It’s not that individuals believe computers or voice assistants are actually human with genders, according to Mayhorn. Instead, it illustrates how people unconsciously introduce their cultural biases and perceive computers as social entities, interpreting them through a gendered framework.
Tech companies’ marketing strategies and user interfaces have perpetuated these effects, as most well-known voice assistants were assigned feminine names and utilized speech patterns derived from recordings of female voice artists. While many tech companies have recently eliminated explicit labels like “female” and “male” from voice options, most popular voice assistants were initially presented as female within their systems, referenced as female by corporate spokespeople, and even programmed to react to gendered harassment with flirtatious responses.
“For our goals—creating a useful, supportive, and trustworthy assistant—a female voice was the best option,” a Microsoft representative told the Wall Street Journal in 2017. Other companies have been less transparent about their rationale for selecting female-coded voices to perform tasks such as scheduling, managing correspondence, and issuing reminders—tasks commonly attributed as female roles both at home and in the workplace.
“In the end, what individuals seek is what boosts engagement,” Zirau articulates.
Is a gender-neutral future possible? It could be. Although most voice assistant technologies currently provide both male and female sounding options, gender-neutral voices are still not widely used, despite various ideas for genderless voice models. An exception is Apple, which has begun to categorize voices in a non-gendered way and offers a gender-neutral Siri voice recorded by a member of the LGBTQ+ community. However, this voice isn’t set as the default option on the device, and psychological studies regarding how users respond to and engage with genderless voices are still in the early stages.
What is evident is that intelligent voice assistants, regarded as quite innovative just ten years ago, are now more prevalent than ever. In 2022, an estimated 142 million individuals were using voice assistants, and this number is anticipated to grow to 157.1 million users—almost half of the U.S. population—by 2026. The COVID-19 pandemic, along with the increasing use of smartphones, accounts for the remarkable growth in these assistants. Mayhorn’s research indicates that older adults are also engaging with voice assistants more than before. “Many of them are more familiar with voice assistance than younger college students,” he points out. “This suggests that we’re witnessing a broader adoption of this technology in households, and that people are beginning to recognize [its] value and significance.”
Considering the apparent saturation of the technology—over half of all Americans are expected to use a voice assistant by 2026—it’s easy to overlook that these assistants are still relatively new, with their first widespread introduction occurring only 13 years ago. This implies that there is still an opportunity for companies and society to confront the potential ramifications of gender stereotyping and voice assistance. Enhancing gender diversity within the technology sector might also contribute positively. However, only time will reveal whether the expectations established by Siri, Alexa, and their female-voiced counterparts—propelled by a society reluctant to abandon the gender binary—can ever truly be challenged.
Siri, Alexa, Cortana, Google Assistant. What do these voice assistants share in common? Well… they are predominantly represented by female voices.
You might not have noticed this before, and while some may contend that many AI voice systems now offer a male voice option, the reality is that the default settings for most voice assistants feature a female voice, typically accompanied by a feminine name.
This type of AI has become increasingly integrated into our everyday lives. Nearly 3 billion people are presently using voice automation technology for tasks such as waking up, finding local restaurants, or checking weekend weather forecasts, and this number is anticipated to grow significantly in the future.
While utilizing platforms like Alexa has numerous advantages, the developers of these voice assistants are facing pushback for preferring female voices, as this may unconsciously reinforce outdated stereotypes and biases that suggest women are subordinate, quiet, courteous, and primarily meant to “assist” others.
So, why is there such a tendency among many companies to associate AI and voice automation with femininity?
In this blog post, we will explore the reasons behind the prevalent gender bias in AI, the difficulties programmers encounter in developing male voice assistants, and potential solutions for the industry to modify this bias.
Factors Contributing to Female Voice Assistants
Preference for Female Voices
Numerous studies indicate that people generally favor the sound of a female voice, with some suggesting that our affinity for female voices might even start in the womb, as these sounds can be soothing and calming.
Additional research has shown that women usually pronounce vowel sounds more distinctly, making them easier to comprehend, especially in professional environments. This observation is not entirely new within the industry.
Historically, female voice recordings were utilized during World War II in aircraft cockpits because their higher pitch allowed them to be more easily recognized than their male counterparts.
However, this preference is often contested, and many misconceptions about the superiority of female voices in terms of clarity through small speakers or amidst background noise have been debunked.
There is also evidence that women may face criticism regarding their vocal nuances. A simple Google search will reveal that if you input “women’s voices are,” one of the top suggestions will finish that sentence with the word “annoying,” which is decidedly unflattering.
Lack of Data for Male Voices
This is a frequently debated point for developers beginning the process of creating automated voice systems.
Historically, text-to-speech technologies have predominantly been trained on female voice samples. Given the large amount of data available for female voices, companies often select them for developing voice automation software because it is the most time-effective and economical choice.
The use of female voice recordings can be traced back to 1878, when Emma Nutt became the first female telephone operator. Her voice was embraced so well that it set the standard for others in the field. By the late 1880s, all telephone operators were female.
Due to this gender shift within the industry, we now possess over a century of female audio recordings that can be utilized to create new forms of voice-automated AI known to resonate with users.
Why invest time and resources in gathering male voice recordings and developing male-voiced AI when the response from users is uncertain? This brings us to the next point…
Challenges in Developing Male Voice Automation
As the AI sector is largely characterized by female voices, it isn’t surprising that constructing male voice automation solutions can be incredibly challenging. For example, consider Google’s experience.
In 2016, Google introduced “Google Assistant,” and there was a reason behind its choice of a gender-neutral name; the tech giant intended to offer both male and female voices for its new assistant.
Unfortunately, the technology that Google relied upon for its assistant was primarily trained using female data, resulting in better performance with female voices.
Google’s older text-to-speech system would assemble audio segments from recordings using a speech recognition algorithm. It functioned by embedding markers in various points of sentences to instruct the system on the beginning and ending of certain sounds.
Brant Ward, who serves as the global engineering manager for text-to-speech at Google, noted that these markers were not positioned as accurately for male voices, complicating the achievement of similar quality for a male voice as it is for a female voice.
Regrettably, the systems that were accessible to Google and other companies at that time had a larger amount of female data compared to male data.
The team behind Google Assistant strongly pushed for the inclusion of both male and female voices; however, the company ultimately decided not to create a male voice after realizing the difficulties involved.
Ward mentioned that developing a male voice for Google Assistant would have required over a year, and even after its completion, there was no assurance that it would be of sufficient quality or well-received by users.
How Can We Address Gender Bias in Voice Automation?
It seems that the gender bias in voice automation stems from insufficient data and widely accepted, yet unchallenged, notions surrounding the female voice. When all this information is considered, the prospect of developing male-voiced automated software may appear insurmountable.
Nevertheless, there are measures we can implement to shift the gender bias not just within voice automation, but across the AI sector as a whole.
1. Invest in Machine Learning Technology
With the advancement of new machine learning technologies, text-to-speech systems are evolving and are now more capable of producing natural-sounding male and female voices for AI.
For instance, Google collaborated with DeepMind, a British subsidiary of Alphabet Inc. and a research laboratory, aiming to develop an innovative type of text-to-speech algorithm that would decrease the number of recordings necessary to create voices that more closely resemble a real human.
By 2017, Google, along with DeepMind, had developed an algorithm called WaveNet, which enabled Google to create more lifelike male and female voices for Google Assistant.
Currently, the American version of Google Assistant is equipped with 11 distinct voices, with various accents. To make its product as inclusive as possible, Google randomly assigns new users one of two primary voices – one male and one female.
2. Establish AI Standards
By 2027, the global AI market value is projected to reach $267 billion. AI, and especially voice assistants, are becoming integral to our daily lives, yet there are no widespread guidelines regarding the humanization of AI.
Many companies that have developed voice assistants continue to choose a female voice and/or a female name, which can reinforce the stereotype of women being in service-oriented roles.
For this reason, we must consider creating and implementing industry-wide standards about how gender is represented in AI. With such standards in place, businesses could develop AI that is more inclusive and gender-balanced.
In creating these new industry benchmarks, it is essential to include input from AI developers and ensure a diverse range of gender identities, sexual orientations, races, and ethnic backgrounds are represented.
With this focus group, we could define the characteristics and proper contexts for “female,” “male,” “gender-neutral,” and “non-binary” human voices.
These industry standards should also consist of a framework for developing unbiased text-to-speech algorithms that are sensitive to gender and culturally relevant.
3. Transparency in Data Collection
To diminish the gender bias present in voice automation data, companies need to be more open regarding their data practices.
Companies should be encouraged to reveal the demographics of their AI development teams, share research findings including user preferences for voices, and disclose information related to gender-neutral AI voices.
All of this data is crucial in understanding the connection between technology, AI, and gender bias, and in discovering new solutions to this issue.
4. Promoting Inclusivity within the AI Industry
This may seem like a straightforward assertion, but the statistics indicate that our industry is not doing enough to attract individuals of various genders into AI careers. Currently, “women constitute approximately 26% of the workforce in data and AI roles globally, which decreases to only 22% in the UK.”
This percentage drops significantly when considering those in AI who identify as transgender or non-binary. To improve these figures, we must inspire and assist more women and people of diverse genders to explore this field in higher education.
A more diverse AI development workforce can better tackle intricate gender issues before and during the creation of new voice assistants. To achieve greater diversity, we need to establish robust educational foundations that are inclusive for everyone.
We can accomplish this by expanding the range of educational opportunities available to students from secondary school onward, while having female, transgender, and non-binary individuals actively participate in developing AI course materials. If students recognize their representation in their studies, they are more likely to pursue further education.
