Social Research with Gender-Neutral Voices in Chatbots – The Generation and Evaluation of Artificial Gender-Neutral Voices with Praat and Google WaveNet

Research output: Contribution to book/conference proceedings/anthology/reportChapter in book/anthology/reportContributedpeer-review

Abstract

Artificial voice generation enables the creation of gender-neutral voice-based chatbots, such as “Q”, and with this, new possibilities for social scientists to investigate effects of gender in chatbots on users. However, for this research, gender-neutral and matching male and female voices are necessary. This paper aims to match this need by providing a method for the creation of acoustically gender-neutral voices which can serve as a base for further investigation. The method requires less technical knowledge than most other approaches while, at the same time, theoretical assumptions about gender perception and neutrality can be modeled into the voices. Specifically, fundamental frequency and formant frequencies are outlined as major influences and subsequently included in voice production. In addition, an efficient two-step test for assessing gender perception is demonstrated. Applied to the produced voices, the test proved to be successful in determining gender-neutral variants. In sum, the paper shows that the production method and underlying theory are suited to create gender-neutral voices as well as matching male and female variants.

Details

Original languageEnglish
Title of host publicationChatbot Research and Design - 5th International Workshop, CONVERSATIONS 2021, Revised Selected Papers
EditorsAsbjørn Følstad, Theo Araujo, Symeon Papadopoulos, Effie L.-C. Law, Ewa Luger, Morten Goodwin, Petter Bae Brandtzaeg
Pages176-191
Number of pages16
Publication statusPublished - 27 Jan 2022
Peer-reviewedYes

Workshop

TitleInternational Workshop on Chatbot Research and Design
SubtitleCONVERSATIONS 2021
Conference number
Duration23 - 24 November 2021
Website
Location
City

External IDs

Scopus 85124669705
Mendeley 9d08f774-d2b1-356f-a91a-583b90e45ce8
ORCID /0000-0001-6515-9985/work/142245100

Keywords

Research priority areas of TU Dresden

Keywords

  • Gender neutrality, Voice production, Voice-based chatbots