AI-generated Rock Music: the Making Of

Aiva Technologies
3 min readOct 24, 2018

--

Since our previous blogpost describing how we used AIVA to create the world’s first AI-generated album of Chinese music, the team has been hard at work to find ways to train AIVA on new styles of music, and ultimately make it easier for composers to scale their songwriting process.

“ AIVA, play me a Rock song ”

Our database of 30.000 scores is largely made up of classical, symphonic and cinematic scores encoded in MIDI format, but we have recently added several hundreds Rock MIDIs as well. The first logical step was to retrain our existing deep neural network architectures on that new type of data.

Melody is key to a really great Rock track, and at first, we didn’t expect anything super interesting to happen on that front. We did expect to have some cool harmonies that would give a nice starting point to a great Rock song, but nothing that would really stand on its own. Sometimes, it can be hard for our neural networks to capture the mechanics of great melodies in styles that are less rich than classical music.

But our AI music composer AIVA proved us wrong, and we’re happy to say the result was above our expectations! Here’s the recording of a Rock track composed by AIVA, that we call On the Edge:

On the Edge — AI-generated Rock Music composed by AIVA

Building a Rock Solid Foundation

While building a “verse” and “chorus” type of structure typical of Rock music is easy because of its repetitive nature, creating highly effective and memorable building blocks for that structure is the challenge of this genre.

Ensuring repetition is relatively easy, because the variety of different musical structure types that can be found in popular music is quite limited. For example, On the Edge uses a typical structure of “Verse, Chorus, Verse, Chorus, Bridge, Chorus, Bridge, Chorus.”

However, creating a memorable tune is a lot more challenging, but it usually comes down to two things:

  1. Creating a catchy melody for the chorus, which stands as the hook for listeners to remember and sing along with
  2. Ensuring that the instrumentals accompaniment has a pleasing chord progression that can stand on its own and allow the members of a band other than the lead singer to really shine

When we first listened to the MIDI file that AIVA had composed for On the Edge, it was hard not to smile because of the typical rock bass line that started playing, followed by a recognizable yet interesting chord progression and a ear worm melody.

A piano roll view of “On the Edge” on AIVA’s Music Engine (www.aiva.ai)

Turning It Up to Eleven

The raw track did have some weaknesses:

  • The bridge contained a melody, which we decided to remove altogether, to give room for the piece to breath and let the instrumentals shine
  • It did not have any drums: those were manually added during the production process
  • The rhythmic pattern of the verse’s melody was awkward, so we proceeded to using our newly built melody generator to recreate a new melody from scratch, by specifying the chord progression of the verse to condition the creation process.

Brad, our Music supervisor here at AIVA, then proceeded to perform the piece on his guitar and finalize the sound production with a quality mix and mastering.

Overall, we’re very excited about the results achieved with On the Edge, because we want to allow composers to collaborate with AIVA in a wide variety of genres. We also want to keep exploring new ways of breaking down the composition process into multiple components , where new tools like our melody generator could help artists create certain elements of a song that they struggle with the most, while also giving them enough space to innovate in other parts of their creation.

If you’re interested to use AIVA to help you write music, you can sign-up to our upcoming Beta here!

♢ JOIN THE COMMUNITY ♢

--

--

Aiva Technologies

AIVA is the Artificial Intelligence composing emotional soundtrack music, to support the creative workflow of composers. Find out more: www.aiva.ai