Which advanced audio tech do singers use?

12 points·stealthmodeclan··4 comments

Has deep learning made its way into singing?

Not yet. Modern vocals are still recorded and tuned manually using AutoTune and Melodyne.

The vocaloid scene (Hatsune Miku) was big for a while in Japan, but that's fairly simple phoneme synthesis, not ML.

I've heard some good AI demos of ML-based vocal synthesis, but nothing is available commercially yet.

Like speech synthesis, it's easy to end up in uncanny valley - which is acceptable for spoken synthesised speech in many applications, but isn't good enough for music.

I'd argue that Melodyne is pretty advanced. I mean, it can take a polyphonic (multi-voice) audio track and separate the notes so you can manipulate them almost like a MIDI track. That's pretty damned difficult for a computer to pull off.

As a musician, I noticed something interesting: musicians tend to be _extremely_ conservatives. Guitarists, for example: try telling them to use a solid state (transistor) amp, and wait for the instant explanation of why valves are absolutely better. Only very recently have "modelling" digital amps (axe-fx) begun taking place in professional productions, and those are quite simple (compared to ML) DSP processes.

One thing I've found to be a big step to using the modelling products is the user experience. Guitarists want to make music not fiddle with menus and software. A physical amp has all the options visible and they are simple. One knob per feature is much more intuitive and faster to learn and use than one that changes it's behaviour per patch etc. So what is described as conservative could just be a shortcoming of modern products' interface.

Reply on news.ycombinator.com