Hear it.

Twenty singers. Nine women, eleven men.

Trained on VocalSet, an open set of recordings anyone is allowed to train on. They sound natural - the pack to reach for when the choir is meant to sound like real people.

9 women · 11 men

• Ada

• Bryony

• Cleo

• Delphine

• Esme

• Freya

• Greta

• Hana

• Ilse

• Abel

• Boaz

• Caspar

• Dorian

• Emrys

• Fenn

• Gideon

• Hollis

• Ivo

• Jonas

• Kester

Eight voices. Four women, four men.

Bright, feathery voices for airy textures and light background layers. They were blended, trained, rendered and retrained, briefly each time, until none of them sounds like anyone in particular. That leaves the renders soft and slightly lo-fi, with a pleasant grain of artifacts.

4 women · 4 men · trained on a mix of VocalSet, session recordings, friends' iPhone recordings, samples and AI-generated vocals

• Monet

• Renoir

• Edgar

• Pisarro

• Morisot

• Cassatt

• Bracquemond

• Lilla

Fourteen voices off 78s. Seven women, seven men.

Soloists who recorded for Victor between 1902 and 1920, from vaudeville tenors to opera stars. We lifted each voice off its record, away from the band, and trained a singer on it. They sing back the way the records sound: narrow, warm and a little far away.

7 women · 7 men · trained on public-domain acoustic recordings, 1902–1920, from the Library of Congress's National Jukebox

• Corinne

  1904–1909

• Edith

  1903–1911

• Ernestine

  1907–1917

• Frances

  1910–1919

• Johanna

  1906–1917

• Marion

  1916–1919

• Olive

  1913–1918

• Billy

  1902–1920

• Byron

  1903–1917

• Irving

  1914–1915

• John

  1910–1919

• Pasquale

  1911–1913

• Pol

  1904–1908

• Reed

  1909–1915

Ten instruments that play what you sing.

Woodwinds, brass and strings, each trained as if it were a singer. Sing a line and it comes back on a flute, a trumpet or a cello, with your phrasing and your timing. Put a few of them together and the choir is a section.

Trained on the Good-sounds dataset (Romaní Picas, Parra Rodriguez, Dabiri and Serra, 2015), Music Technology Group, Universitat Pompeu Fabra, used under CC BY 4.0.

Nine voices. Five women, four men.

Nine singers from very different sources: pop singer-songwriters, indie-rock frontman, a J-pop voice, a German vocalist and lo-fi singers. The mismatch is what makes the choir sound rich.

5 women · 4 men · trained on four open singing datasets

• Owen

• Felix

• June

• Theo

• Kenji

• Nora

• Elise

• Iris

• Maya