DSP (Digital Signal Processor) is used anywhere where a digital signal is processed like audio, video, etc. When you play your favourite media its played by your processor’s DSP instead of your CPU saving battery. Speech recognition is another area where DSP is used for this.

Nowadays, it does more than just play media. Including doing AI tasks on a NPU(Neural Processing Unit) like Object recognition, running LLM(Large Language Models) to generate pictures, suggest frequently used apps, etc.

and what kind of code can take advantage of this? apps? javascript in browser apps? or a non-app system process with a specific privilege?

As for code anything that processes signals can be accelerated by it.

User code does not get privileged access to it. JavaScript is sanboxed but system processes in chrome and firefox can use it for media playback.

For accelerated AI tasks on the NPU. It depends if the app developer leverage the specific neural SDK for Qualcomm, mediatek. Or use NNAPI API, or LiteRT

It’s standard on most smartphones like the CPU, GPU. If you want you can ask perplexity.ai for specific info in it.

I have given a short summary. But, there’s lot more you can read if interested.