Turn on the radio to listen!
Raise Your Voice is an art project addressing the ‘authority gap’ between age groups. By transforming any voice to that of an adult, it empowers juveniles to speak up, unhindered by the biases that usually undermine their authority and influence.
While the youngest generation has important ideas, minors are generally taken less seriously than adults, and struggle to be heard and considered. Changing societal biases is too slow, so Raise Your Voice provides a tongue-in-cheek shortcut: It levels the playing field for children and juveniles by converting their voice into that of an adult. Speakers of any age record statements and dialogues at the installation, and choose to listen to the converted recordings in private or broadcast them to radios distributed on site. Decoupling speakers from utterances allows listeners to reflect on how they perceive what was said, and uncover associated biases. Raise Your Voice teaches us to listen more carefully to those who are not currently in charge, and who will thus make the biggest difference for the future.
Raise Your Voice is exhibited at the Ars Electronica Festival 2024, as part of the Linz Institute of Technology Exhibition. It is located on the first floor of the library building at JKU MED Campus, Krankenhausstraße 5, Linz, Austria.
The exhibit runs from Tuesday, September 4 to Sunday, September 8, 2024.
From September 4—9, just using your browser, you can convert your own voice to that of an(other) adult or to that of a child. Try the Web App!
Visitors on site can decide to broadcast their converted statements and dialogs. To listen, tune in to FM 103.2 MHz in or near the library building at JKU MED campus, or to the web stream below. Take in how you can judge statements only by their content, not knowing the speakers' age or gender.
Turn on the radio to listen!
We set up a prerecorded collection of adult voices, each around 60 seconds, taken from a public dataset or by volunteers (you are welcome to visit us at the exhibition to donate your voice!). To convert a new recording to one of these voices, the recording is split into tiny segments, and each of these segments is replaced by a similar-sounding segment from the target voice. The key trick is that this comparison and replacement does not operate directly on the audio waveform — this would sound terrible —, but in a feature space where similar phonemes are close together (using a neural network called WavLM). After replacing each segment by one from the target recording, it is converted from that feature space back to audible sound (using another neural network called HiFi-GAN). The full process was proposed and published as knn-VC. While we did not invent knn-VC, we worked on voice conversion before at our institute, and selected it among several alternatives for its elegance and how well it works with children's voices.
