"A voice is inherently communal. I learned how to use my voice by mimicking the people around me through language, through centuries of evolution on that, or even vocal styles. A pop music vocal is often you're kind of emulating something that came before and then performing your individuality through that kind of communal voice. So I wanted to find a way to kind of reflect that communal ownership" - experimental musician Holly Herndon says to The Fader about her audio deep-fake AI Holly+. Herndon encourages her fans to upload audio files so they can be sung in her voice. She goes into the metaphysics of it: "I mean, we've been able to kind of re-animate our dead through moving picture or through samples, but this is kind of a brand new kind of field in that you can have the person do something that they never did. It's not just kind of replaying something that they've done in the past. You can kind of re-animate them in and give them entirely new phrases that they may not have approved of in their lifetime or even for living artists that they might not approve of. So I think it opens up a kind of Pandora's box".

Berlin-based musician and sound artist Holly Herndon has released a new artificial intelligence tool Holly+, which she refers to as her “digital twin”, that allows fans to upload any polyphonic audio and receive a new version of that music sung in Herndon’s own voice. Holly+ is as much a technological and artistic experiment, as it is a response to, and embrace of, the rise of deepfake technology, The Fader reports.

“As a piece of engineering, it’s really impressive” - an electronic musician and academic dr Matthew Yee-Kin says to the Guardian about audio deepfakes - “they break down an audio signal into a set of lexemes of music – a dictionary if you like – at three different layers of time, giving you a set of core fragments that is sufficient to reconstruct the music that was fed in". The G in convinced, rightly so, deepfake music is set to have wide-ranging ramifications for the music industry - any company that wouldn't want to pay the market rate for using an established artist’s music, they could create their own imitation. And big companies are getting into it: Google is working on it within their Magenta Project; startup Amper Music is producing custom, AI-generated music for media content; Spotify employed François Pachet, former head of Sony Music’s computer science lab, in its AI research group...

Stereogum jumps into the audio deepfakes wagon discussion with a long read about human creativity and the future of music generated by artificial intelligence. The story kicked off on April 30 when audio deepfakes were made public by an independent research organization out of San Francisco called OpenAI. It took recordings of real artists - Jay-Z, Elvis, the Beatles, and thousands of others - a total of 1.2 million songs, entered them into its open source model powered by a neural network, which was then given various levels of guidance and instructed to generate new works. The artificial intelligence finally made some deepfakes. General Intelligence and Pitchfork already published their takes on the story.

What will the future bring for the audio deepfakes - AI-generated imitations of human voices, Pitchfork tries to find answers in an analysis of Jay-Z deepfakes. YouTube channel called Voice Synthesis posted several Jay-Z deepfakes in April - Shakespeare’s 'To Be, Or Not to Be' soliloquy from 'Hamlet', Billy Joel’s 'We Didn’t Start the Fire', and a decade-old 4chan meme. They were removed two days after due to a copyright claim, but soon they returned. The software has to be “trained” with audio samples and text transcripts, and the actual voice is used in the creation but from there it’s all ones and zeros from the AI, so it's a legal grey zone. Musicians and fans could potentially be grappling with the weird consequences of AI voice manipulations long into the future - the P is guessing, with high probability actually.

A great article by General Intelligence about music generated by artificial intelligence, so good it's nearly impossible to tell the difference. What these algorithms do is they analyze large collections of an artist’s songs, find patterns in the audio data that humans would correlate with hallmarks of music style, and then use those patterns to generate new audio. One such is Jukebox, an algorithmic system able to generate music - complete with lyrics - in the style of famous musicians like Elvis and Frank Sinatra.