Breaking the Binary: A Systematic Review of Gender-Ambiguous Voices in Human-Computer Interaction

Research Background and Issues

• Issues and Challenges: Studies have indicated that gendered voice systems (e.g., voice assistants or robots) may reinforce gender biases and stereotypes, particularly by defaulting to female voices for assistant roles. In recent years, societal attention to gender diversity and inclusivity has driven researchers to develop gender-ambiguous voices to reduce bias and broaden user experiences. However, there is currently limited understanding of the definition, creation, and user perception of gender-ambiguous voices.
• Significance: Voice interfaces are widely used in daily life, and their design choices can profoundly impact the interaction between technology and social behavior. Moreover, defaulting to gendered voices may alienate user groups with more complex gender identities.
• Research Motivation and Related Work: Drawing from gender theory and linguistic studies (e.g., Judith Butler's theory of gender performativity), it is recognized that gender is a social construct. The motivation for this research is to explore whether gender-ambiguous voices can reduce gender bias, promote technological inclusivity, and open new directions in the field of human-computer interaction.

Solutions

• Proposed Approach: This paper systematically analyzes research on gender-ambiguous voices in the field of human-computer interaction, summarizing 36 studies with a focus on their definitions, characteristics, creation methods, user perceptions, and evaluation approaches. Additionally, the authors propose the following specific action points:
  1. Clarify terminology (e.g., gender-ambiguous, gender-neutral, non-binary).
  2. Conduct acoustic analyses of gender-ambiguous voices.
  3. Develop standardized metrics for evaluating gender-ambiguous voices.
  4. Establish an open-source repository for gender-ambiguous voice resources.
  5. Create frameworks for the development and application of gender-ambiguous voices.
• Innovative Contributions:
  • Proposes a unified definition of terms, such as adopting "gender-ambiguous" as the preferred term to avoid misleading expressions like "gender-neutral" or "genderless."
  • Provides a preliminary acoustic analysis framework to identify key vocal features (e.g., pitch, resonance, vocal weight).
  • Designs standardized evaluation methods (e.g., a five-point Likert scale) to measure user perception of gender-ambiguous voices.
• Implementation Steps and Key Technologies:
  • Conduct systematic searches using two major literature databases (ACM Digital Library and IEEE Xplore) to identify relevant studies.
  • Extract corresponding vocal features from the research and analyze them using acoustic software (Praat).
  • Create or modify voice characteristics using developed AI technologies (e.g., Tacotron 2, Google WaveNet).

Research Outcomes

• Specific Findings:
  • Analyzed existing gender-ambiguous voices and extracted acoustic data (e.g., pitch, speech rate, jitter values) from four samples.
  • Proposed an aggregated definition of terms related to gender-ambiguous voices, laying the foundation for a unified research language.
  • Compared existing voice generation methods, including commercial TTS systems (e.g., Amazon Polly) and custom voice generation technologies, highlighting the complexity of creating gender-ambiguous voices.
• Advantages Over Existing Solutions:
  • Emphasizes the use of "gender-ambiguous voices" rather than completely eliminating gender characteristics, aligning better with users' diverse cognitive habits.
  • Provides open resources (e.g., open-source voice libraries) to enhance research reproducibility.
  • Offers practical evaluation frameworks to ensure inclusivity and usability in voice design.
• Experimental or Evaluation Results:
  • User perception experiments revealed that when voice characteristics incorporated ambiguous features like pitch and resonance, some users perceived these voices as more "ambiguous," though this perception was influenced by users' gender and cultural backgrounds.
  • Results showed that non-binary groups had a higher acceptance of ambiguous voices, particularly those with more feminine characteristics.
• Limitations and Future Directions:
  • This study analyzed only English voices, potentially overlooking cross-linguistic and cross-cultural differences.
  • The sample size was small (only four voices analyzed), and future research should expand the scope to improve generalizability.
  • Future work should focus on developing more open-source voice libraries and exploring user perceptions of non-gender-related characteristics of ambiguous voices, such as friendliness and trustworthiness.

Conclusion

This paper systematically reviews the state of research on gender-ambiguous voices in the HCI field, laying a critical foundation for this emerging area and providing directions for future research. By clarifying terminology, proposing an acoustic feature framework, and designing evaluation standards, the study advances the creation and application of gender-ambiguous voices to promote technological inclusivity and social equity. These findings are significant for developing more representative and diverse human-computer interaction technologies.