We’ve all see the explosion of AI song writing (awful composing, but that’s not the point here). Though the composing is bad, what can be stunningly good are the actual voices/instrumental sounds that are created. Here’s a classical-style example, again the AI composing is wonky, but the AI sounds are absolutely amazing, probably more lifelike than any sound library I’ve ever heard:
I imagine this would make EastWest/Native Instruments/etc, feel very nervous. It seems like it will only be a matter of time before you can upload a score to a service that could play and sing it quite well and the whole painstaking process of sequencing with gigantic sound libraries will mostly become a forgotten art.
My question then is, is any work being done on AI powered instrumental playback libraries, or on services that will simply read a PDF of a score and create a convincing AI performance? It would seem to me like whichever scoring software does this first, will take the world by storm.
I agree with you wholeheartedly on that one. The part that impressed me is the fabulous quality of the orchestral sounds, the realistic enunciation, the voice quality, etc. The composing is a hodgepodge of good and bad melodic lines woven together in an odd fashion, but the playback is truly awesome. What would it be like if Dorico allowed us to playback a notated piece of our own using an engine this good? Or to upload a PDF and get it “performed” in this way. The foundational technology is obviously already out there. Who will be the first to make use of it in notation software?
I think that one of the reasons NotePerformer decided to not continue their Playback Engines (which I still use in version 4.5.1, it sounds great with BBCSO), is because they anticipate that AI sound quality will continue to improve. Having NotePerformer’s musical intelligence with really excellent sounds, as good or better than the best sample libraries, is an exciting possibility.
Perhaps, but I understood that he ran into licensing issues with one (or more) of the library creators. For a while I was quite impressed with the Opus engine and used it regularly, but then I opened an older score with my custom template and it actually sounded better in many ways (some ways not, obviously). But that “Noteperformer” logic coupled with professionally sampled libraries really does bring an arrangement to the next level.
Yes, you’re right - the licensing issue was perhaps the main reason to discontinue the Playback Engines. But I think they also said something about future possibilities for AI. Which of course could be bad news for the sample library companies. I’m glad Arne et al. seem to be thinking ahead - for me they’re a model of an excellent company.
The part that impressed me is the fabulous quality of the orchestral sounds, the realistic enunciation, the voice quality, etc. The composing is a hodgepodge of good and bad melodic lines woven together in an odd fashion, but the playback is truly awesome. What would it be like if Dorico allowed us to playback a notated piece of our own using an engine this good? Or to upload a PDF and get it “performed” in this way. The foundational technology is obviously already out there. Who will be the first to make use of it in notation software?