Reference Voice
2021
Three-channel sound installation, score
16 min
Dimensions variable
Why are most of the automated voices we hear female? In this work, I investigate the intersection of voice automation technology and gender. Researching the close relationship between the two, I came up with some "conditions," including physical traits commonly attributed to feminine voice, stereotypes around it, and critical interpretations of femininity and technology. I then commissioned a composer to create a composition that reflects these conditions.
The resulting piece is performed by a non-female singer, who attempts to mimic the female voice. In addition to standard singing instructions, the composition includes gender-ambiguous sounds such as click consonants (i.e., "tsk tsk") and sibilant. The composition and the singer's assimilation exemplify the performativity of gender—how gender is performed in proximity to given norms. The work brings awareness to these norms and conditions that frame femininity.
<Things to consider for composition>
1. The vocalists’ range should be between 180 Hz to 250 Hz, which is the typical range of a female speaking voice.
2. AI assistants developed in South Korea are mostly given a female voice by default. Developers tend to choose a cheerful and amiable voice of a female in her 20s—the vocalists should aim to emulate these vocal characteristics.
3. In 2016, Microsoft launched a chatbot named Tay that identified as a teenage girl. Tay was shut down only sixteen hours after its debut. The chatbot was designed to learn human language openly through texts and tweets on the Internet, but was inadvertently taught the language of white supremacy, misogyny, and anti-Muslim hate speech. As a result, Tay blindly posted hate speech toward feminist groups and POCs, leading the developer to quickly terminate the program. The length of this piece is sixteen minutes as a tribute to Tay’s brief existence.
4. In 1996, Bradlow published a study that female voices can be more easily understood due to gender differences in oral structure, allowing females to pronounce richer and more exact vowel sounds. The results of studies like this often serve to provide a biological basis for electing to assign a female voice and identity to machines. In 2019, Sophia, an AI robot, was asked during an interview: “Do you regard yourself as male or female?” to which she replied, “Female.” The interviewer asked again: “Why do you think you are female?” Sophia’s response: “I'm a robot so technically I have no gender, but identify as feminine and I don't mind being perceived as a woman.”
The all available vowel sounds excerpted from Sophia’s answer above provide the lyrics for this piece: A, E, I, O, and U.
5. It is hard to distinguish the gender of someone who is whistling if listening only to their whistle. The same goes for tongue clicking. Unless it is a speaking or singing voice, it is impossible to assign a gender to most sounds that are produced with the mouth. In this piece, the vocalist should produce diverse sounds other than speech or singing (e.g. whistling, tongue clicking, kissing, sucking through teeth, etc.)
6. The voices heard in this piece may not sound feminine or they may be heard as forcedly feminine. Some may regard that vocalists’ attempts represent the current sonic world of AI technology, which is not gender-neutral and has tended to develop machines that reinforce the metaphor of female presence. Others may consider that this piece reflects the way machines learn and extend human bias, and respond in a human voice. Still others could perceive that the voices present a clumsy human version of "Q," which is the genderless voice created in 2019 to counter AI gender bias.
This work was commissioned by Seo-Seoul Museum of Art in Seoul in 2021
(KR)
2021
Three-channel sound installation, score
16 min
Dimensions variable
Why are most of the automated voices we hear female? In this work, I investigate the intersection of voice automation technology and gender. Researching the close relationship between the two, I came up with some "conditions," including physical traits commonly attributed to feminine voice, stereotypes around it, and critical interpretations of femininity and technology. I then commissioned a composer to create a composition that reflects these conditions.
The resulting piece is performed by a non-female singer, who attempts to mimic the female voice. In addition to standard singing instructions, the composition includes gender-ambiguous sounds such as click consonants (i.e., "tsk tsk") and sibilant. The composition and the singer's assimilation exemplify the performativity of gender—how gender is performed in proximity to given norms. The work brings awareness to these norms and conditions that frame femininity.
<Things to consider for composition>
1. The vocalists’ range should be between 180 Hz to 250 Hz, which is the typical range of a female speaking voice.
2. AI assistants developed in South Korea are mostly given a female voice by default. Developers tend to choose a cheerful and amiable voice of a female in her 20s—the vocalists should aim to emulate these vocal characteristics.
3. In 2016, Microsoft launched a chatbot named Tay that identified as a teenage girl. Tay was shut down only sixteen hours after its debut. The chatbot was designed to learn human language openly through texts and tweets on the Internet, but was inadvertently taught the language of white supremacy, misogyny, and anti-Muslim hate speech. As a result, Tay blindly posted hate speech toward feminist groups and POCs, leading the developer to quickly terminate the program. The length of this piece is sixteen minutes as a tribute to Tay’s brief existence.
4. In 1996, Bradlow published a study that female voices can be more easily understood due to gender differences in oral structure, allowing females to pronounce richer and more exact vowel sounds. The results of studies like this often serve to provide a biological basis for electing to assign a female voice and identity to machines. In 2019, Sophia, an AI robot, was asked during an interview: “Do you regard yourself as male or female?” to which she replied, “Female.” The interviewer asked again: “Why do you think you are female?” Sophia’s response: “I'm a robot so technically I have no gender, but identify as feminine and I don't mind being perceived as a woman.”
The all available vowel sounds excerpted from Sophia’s answer above provide the lyrics for this piece: A, E, I, O, and U.
5. It is hard to distinguish the gender of someone who is whistling if listening only to their whistle. The same goes for tongue clicking. Unless it is a speaking or singing voice, it is impossible to assign a gender to most sounds that are produced with the mouth. In this piece, the vocalist should produce diverse sounds other than speech or singing (e.g. whistling, tongue clicking, kissing, sucking through teeth, etc.)
6. The voices heard in this piece may not sound feminine or they may be heard as forcedly feminine. Some may regard that vocalists’ attempts represent the current sonic world of AI technology, which is not gender-neutral and has tended to develop machines that reinforce the metaphor of female presence. Others may consider that this piece reflects the way machines learn and extend human bias, and respond in a human voice. Still others could perceive that the voices present a clumsy human version of "Q," which is the genderless voice created in 2019 to counter AI gender bias.
This work was commissioned by Seo-Seoul Museum of Art in Seoul in 2021
(KR)
#1-5 Installation View at Nam-Seoul Museum of Art, Seoul (Photo by David Cardonal)
2 min 42 sec Video Documentation