From the deep fake memes of Eminem explaining music theory concepts to Gordon Ramsay to using Chat GPT to create poems to set to music, AI is here to stay and will have a palpable impact on choral music. So often, the conversation around AI in music focuses on ethical questions surrounding copyright infringement, uncertainty about the future, and outright fear of replacement amongst musicians. What we don’t talk about are existing AI tools that can (and will) revolutionize the way we write and share choral music on the internet.
If I compose a string quartet or an organ prelude, I can write it in Sibelius or MuseScore and hear a playback that gives me an audio output that is 70% as good as if I had recorded that music with live players. Digital playbacks are by no means as good as having live performances of your compositions, but for getting recordings for student assignments, university applications, or even for creating social media content, digital playbacks get the job done.
I’ve used most of the mainstream notation softwares to write choral music, but I’ve never seen a playback that incorporates text. Sensitivity to text is fundamentally important to choral music and without a way to create digital renderings that incorporate text, the only way to get good demos of choral compositions is to record them with a choir, which is significantly more difficult than exporting a Sibelius file as an MP3.
There are two main ways for young choral composers to get their work sung: find a choir to sing your work or start a choir to sing your work. Even if you can convince a conductor to program a three-minute song or can get 20 singers for a few hours to read through some of your work, I’ve found that both these routes have ridiculously high barriers to entry, and require a significant amount of time, energy, and focus to realize. Also, writing new choral music faster than what the conductors and singers in your network can workshop or perform poses a problem. The solution: find ways to get convincing digital renderings of choral music that would be 70% as good as if they were sung by a choir with 20% of the work needed to make that happen.
After scouring the internet to find ways to do this, I dived down the rabbit hole of online AI vocal editing tools and discovered some pretty exciting products. My approach was to get recordings of my compositions by singing all four parts in my own octave (I’m a baritone) and use AI to digitally alter my voice to become a soprano, alto, and tenor. (I’ve talked in detail about how to do this from a technical perspective and which AI tools to use on my podcast Regarding Music). I upload an MP3 of me singing and an AI transposes my voice into a different register and applies one of many different stylistic presets to that recording. In short, I can make myself sound like a female jazz singer. After getting AI-altered recordings of my own voice to get my soprano, alto, and tenor parts and record the bass part normally, I cut the four parts together as if I’d done a multi-track recording. Boom - I can record all my compositions on my own, in my basement studio without ever having to leave the house.
It's not that easy. I’d say this method can render a convincing demo that’s 70% as good as if it was sung by a choir, but I had to overcome some pretty significant technical and musical challenges. The first of these was that the AI stylistic presets I used seemed to be designed to make your voice sound like a country, jazz, or pop singer. I think the products I used were geared more towards creating backing vocal tracks on pop and country tracks than on creating a choral blend. Whilst it would have been awesome if I could select a choral pre-set that would make my voice sound like Westminster Abbey, I found that if I combined multiple presets, I could edit them in such a way that had a more traditional choral sound. This was especially the case for creating soprano parts. I used a combination of a Broadway pre-set that had a fair amount of vibrato, a female pop sound, and a really breathy pre-set that I recorded in three separate tracks. This enabled me to have more control of the quality of the voice by adjusting the levels of each of those three vocal sounds in my editing software.
Also, I found the way I sang in the recordings was critically important in this process. The importance of note accuracy in choral singing is a bit of a no-brainer, but I found that it was even more critical in this process. Many of the presets I used had built-in auto-tuners (turn these off if you can, it’s much easier to get a good recording). Being a bit flat in such a way that might be acceptable, but not optimal, in a choir, could make an auto tuner round down to a wrong note. I found that this was particularly difficult with really chromatic lines.
Here are a few more pieces of advice to anyone who might want to experiment with using this process. First, I found that singing with vibrato could really mess things up. It was much better to sing with a clear voice and use the stylistic presets to stylize these recordings.
Second, using different transpositions for different vocal parts is the way to go. When I started, I just sang the alto and soprano parts down an octave. This worked pretty well for soprano parts, but the alto parts could get low in my range. I’ve found that doing soprano parts an octave lower, alto parts a fifth lower, and tenor parts a third lower usually works well for me but this will depend on your range and the range of the specific part you are working with.
Third, editing the parts together is much easier if you record using a metronome. I usually sing along with the playback in my notation software. This guarantees that my parts will line up well in the editing phase and makes it faster and easier to learn the parts.
Finally, start off with something easy. My first stab at this was with one of my compositions that was fairly difficult to sing. I found that I spent most of my time learning the parts and getting a good recording rather than working with the AI tools to create a choral blend. The second recording I did was of an easy hymn that I know like the back of my hand.
I wouldn’t say that using AI to record compositions yields better recordings than if they were sung by a choir. However, what this process offers is an actionable way for choral composers to create demos of their work to use in applications, to create content, and to mess around with for personal use. A choral demo like this could also offer a more compelling submission than a score alone to a conductor who might consider performing or workshopping one of your compositions. The AI tools and artistic process I’ve found to work for recording my compositions haven't replaced choir and I would much rather have choirs perform my work. AI tools like the ones I’ve used do not offer a replacement to choir, but rather something new in the toolkit for choral composers who might not have otherwise have the means to share their music.
*Opinions expressed on The CCCC Blog are reflections of the individual author, and may not represent all members of The CCCC Community.
Comments