Synthetic voices in AI assistants reflect deep-rooted gender stereotypes

Digital voice assistants now manage alarms, answer questions, and mediate household routines for millions of users, often through voices that sound distinctly female. These systems are often presented as efficient tools designed to maximize usability and naturalness. A new study challenges that assumption by framing synthetic voices as ideological artifacts

In Gender by Proxy: Echoes of Femininities in Synthetic Voice Design, published in AI & Society, author Carina Lozo argues that synthetic voices such as Amazon's Alexa and Apple's Siri are not neutral tools but engineered performances of femininity, with measurable acoustic differences that reflect domestic and professional gender roles

Synthetic voices as ideological artifacts

According to the study, synthetic voices are not spontaneous expressions but products of sociotechnical design processes shaped by market logic, cultural norms, and entrenched beliefs about gender.

Voices carry social meaning beyond linguistic content. Decades of research in sociophonetics have shown that pitch, breathiness, and phonation style influence how listeners perceive traits such as authority, warmth, competence, and trustworthiness. Lower pitch is often associated with dominance and leadership, while higher pitch and breathiness are linked to warmth, politeness, and traditional femininity.

In synthetic speech systems, vocal parameters are selected, calibrated, and optimized during the design process. As a result, voice assistants do not merely imitate human speech; they encode cultural assumptions about how a helpful or trustworthy assistant should sound.

The paper introduces the concept of performing gender by proxy. Unlike human speakers, voice assistants do not possess embodied identities or lived gender experiences. Their gendered qualities arise from design decisions that translate abstract social norms into acoustic output. Designers, engineers, and corporate stakeholders effectively script femininity into pitch contours, timbre, and vocal texture.

This translation process is rarely visible to users. Instead, the resulting voice is perceived as natural or self-evident. Lozo argues that this invisibility reflects a broader process of fetishization, where the labor and ideology embedded in technological artifacts are concealed, making gendered traits appear inherent to the device itself.

Acoustic evidence of gendered design

The analysis measures three core voice parameters: fundamental frequency, spectral tilt, and harmonics-to-noise ratio. Together, these dimensions provide insight into pitch, breathiness, and vocal clarity.

The results reveal statistically significant differences between the two assistants across all three measures. Alexa's voice exhibits a higher average pitch than Siri's. It also shows a higher harmonics-to-noise ratio, indicating a clearer and smoother sound quality. Most notably, Alexa's spectral tilt values indicate a markedly breathier voice.

Breathiness is widely associated in sociolinguistic research with warmth, softness, and intimacy. In many cultural contexts, it is linked to traditional femininity and affiliative behavior. Alexa's acoustic profile therefore aligns with a stylized, idealized form of domestic femininity. This design mirrors Alexa's primary positioning as a home assistant integrated into kitchens and living rooms.

Siri, on the other hand, displays a lower average pitch and a more creaky phonation pattern. Its harmonics-to-noise ratio is lower, producing a comparatively flatter and less resonant sound. Creaky voice and lower pitch have been linked in prior research to perceptions of authority, professionalism, and competence.

The author interprets Siri's acoustic profile as indexing a more modern, professional femininity. As a mobile assistant tied to productivity and work-related tasks, Siri's vocal qualities align with a persona that signals efficiency and expertise rather than domestic warmth.

These distinctions demonstrate that gendered voice design is not monolithic. Even within female-coded assistants, multiple femininities are engineered. The domestic helper and the professional aide are sonically differentiated, reflecting broader social hierarchies of gender and labor.

The study also acknowledges that some acoustic features may arise partly from technical constraints of speech synthesis systems. However, the study emphasizes that unintended artifacts can still carry social meaning. Users interpret vocal cues through culturally shaped expectations. Whether deliberate or emergent, the resulting voice participates in reproducing gender norms.

Anthropomorphism, market logic, and domestication

The research places its findings within broader debates in human–computer interaction and science and technology studies. One key framework is the Computers Are Social Actors paradigm, which shows that people instinctively respond to computers using social norms such as politeness and friendliness. When voice assistants are anthropomorphized, users attribute personality traits and social identities to them.

This social treatment reinforces the ideological impact of synthetic voice design. When a feminized assistant consistently responds in a compliant and pleasant tone, users may unconsciously align that voice with traditional expectations of female service roles.

The study connects this dynamic to the concept of domestication. As voice assistants migrate into private homes and daily routines, their repeated presence normalizes their vocal traits. Over time, their breathiness, pitch, and scripted helpfulness become familiar and unremarkable. The ideological labor behind their creation fades from awareness.

The paper also highlights how user-centered design can inadvertently reproduce bias. Designers often assume universal user preferences, including the belief that female voices are more pleasant or approachable. Yet research shows that voice preference varies by context and is shaped by gender stereotypes. By defaulting to feminized voices, companies may reinforce assumptions about women's roles as supportive and service-oriented.

The study further addresses the issue of abuse directed at feminized voice assistants. Prior research has documented that systems with female-coded voices receive higher levels of verbal aggression. Early iterations of assistants sometimes responded to harassment with flirtatious or playful remarks, reinforcing expectations of compliance. Although recent updates have introduced more assertive responses, the underlying gender coding remains largely intact.