[Csnd-dev] ANNOUCEMENT: New Csound7 plugin opcodes for embedding AI in Csound pieces
Michael Gogins <[email protected]>
| Newsgroups | gmane.comp.audio.csound.devel |
|---|---|
| Message-ID | <CAFswGP2SC2DcB0N9enk-3V77iyyLV1x0qN=oNV7Wg2Jd9dRxhQ@mail.gmail.com> |
Native builds for macOS, Linux, and Windows are included in the latest release of csound-ac <https://github.com/gogins-dev/csound-ac/releases>. These opcodes enable users of Csound 7 to embed prompts to artificial intelligence models as natural language in .csd files, and to receive responses to the prompts synchronously or asynchronously as Csound data in various formats including strings, arrays of strings, numbers, arrays of numbers, and compiled Csound instrument definitions. For more information, see the README.md <https://github.com/gogins-dev/csound-ac/tree/master/modelprompt>. For an example of a piece that prompts for and runs a new Csound instrument definition, and prompts for and performs a generated Csound score, see this <https://github.com/gogins-dev/csound-ac/blob/master/modelprompt/examples/anthropic_sonnet_instr_score.csd> . This work was inspired by a demonstration of a similar facility by Thorin Kerr in last week's International Csound User Group meeting. I designed the opcodes with reviews by ChatGPT and corrections by me. I used Cursor to generate all of the C++ code. The modelprompt opcodes are a work in progress. Bug reports and suggestions are welcome here <https://github.com/gogins-dev/csound-ac/issues>. ----------------------------------------------------- Michael Gogins Irreducible Productions http://michaelgogins.tumblr.com Michael dot Gogins at gmail dot com