From subtle pitch correction to over-the-top T-Pain madness, autotune is among the most ubiquitous vocal effects in modern music. But what if an AI could detect whether a track uses the effect before you’ve even hit play? Well, after a significant amount of work, we have trained our AI to do exactly that. Here’s how the new autotune classifier works, and how it can help with A&R, sync and playlist curation.
When people talk about autotune, they can mean two different things. First, there’s pitch correction, which is intended to transparently “fix” wayward pitches. It’s used by the vast majority of singers, both good and bad. When used well, it makes singers sound naturally in tune.
Then, there’s stylistic autotune. This is characterised by an immediate and scale-locked retuning of vocals. Think T-Pain or Future.
There are four values that the new autotune tag can have.
None: The vocal is raw and untuned. You’d tend to find this type of vocal in genres such as folk, or in live, acoustic or demo recordings.
Low: There is some form of pitch altering effect, but with low presence. The effect is subtle and might not be anything you can actively hear, but the algorithm has detected that the vocal has been tuned.
Medium: This can mean a couple of things. Either the vocal tuning is clearly audible and not stylistically applied – it’s part of the vocal processing. Or it can mean that there is audible stylistic use of autotune, but it’s not present for the majority of the track. For example, you may have a heavily autotuned chorus, but less processed verses. This result is because the algorithm not only detects the degree of autotune applied, but also the duration that it’s present in a track.
High: Audible autotune is the vocal sound. It’s a stylistic choice and present for most of the track.
The autotune classifier offers a brand new path into catalogues. If a music supervisor is looking for a track with prominent autotune – or none – they now have this data to get to relevant tracks more quickly.
Head of Music Olivia Coléon explains: “Imagine you’re searching for a Lil Wayne-style track. You can filter the autotune classifier by ‘high’ values to find more of the tracks you want, making search within a genre such as trap even more precise.”
For A&R teams, there may be opportunities to discover new raw vocal talent by searching only for tracks with no autotune. Or, for labels looking for tracks that use autotune as a stylistic effect, they can filter by the Medium and High values.
Commercial Director Mack Hampson says: “Ultimately, the more insights catalogue owners and A&R teams can get from music to make sense of the vast number of tracks being created every day, the better. The autotune classifier is another tool to help users navigate tracks quicker and help A&R teams discover great talent.”
For playlist curation
No curator is getting away with a generic ‘hip hop’ playlist in 2022. The genre is as broad as it is popular. So, finding a niche is key. Say you’ve already filtered by low BPM, medium energy, and mood. This is a great start. The new autotune classifier offers another way to refine the sound of your playlist.
For example, for an old-school style “Allergic to Autotune” playlist, you could exclude anything with a Medium or High Autotune value.
Try it for yourself
Any time we release a new product, it goes through significant testing. We will only release new classifiers for Musiio Tag with an accuracy score of over 90 per cent.
Autotune is now available in our Tag Demo and Musiio Tag products. Use it, try to break it, and let us know if you get any surprising results!