task: generate/mix digital audio file (mp3) from referenced audio files(samples) and a song definition file.
implement on server as on-demand (per call) background task and without audio-hardware(no realtime recording) on linux server.
what exists: digital audio step-sequencer you can check out at [url removed, login to view] - german language - which generates songs as serialized PHP data in a mySQL database that consists of:
a) 7 tracks with global settings including: "mute", "solo", "volume" and "pan"
b) songs are built of max 16 different patterns and a pattern sequence of max 20 patterns defining the order in which the patterns are played.
c) patterns consist of 32 positions with zero or following properties: "soundnumber + pitch" - reference an audiofile(sample), "volume" and "pan"
probably a job for sox, php and ...
If you need more info please ask.