Not sure if it’s been discussed yet, but Plex have built this functionality into PlexAmp — under the moniker of ‘Sonic Sage’ — and got round the cost issues by asking users to add their personal OpenAI API key.
That doesn’t feel like a particular user friendly or long term solution, but it’s a great minimal viable product and shows how this sort of natural language discover will likely become a de-facto feature of discover tools / interfaces of all kinds — not just music.
Although I raised the issue of API costs in my previous post, at $0.03 (3 cents) per 1000 words/tokens and the average prompt being around 100 words/token, so 0.3 cents per query, Roon could almost certainly provide a fairly generous monthly allowance to each user without breaking the bank. I expect after the initial excitement most users will use no more than a couple of dozen queries a day, if that.
To some extent I expect absorbing these kinds of costs is going to part of doing business and competing with the competition, as I’d argue it’s almost a given that this kind of text prompt discovery will be built into every streaming services’ native interface within a few years.