sfixpad

After searching for a soundboard that would let me use the sound files I have on my drive, was easy to use, and had a few randomization options to generate non-repetitive soundscape, I realized I'd have to write one myself. 

So here is the first version of "sfix pad". You can add any sound file (including music track) that is in the format mp3, ogg or wav, and let the file repeat after a fixed or random interval of your choosing, and in a volume that, again, can be fixed or random within a range you choose. 

How to use it :

Just click 'NEW' to add a module/sound, click anywhere (that is not a control element) in the box that just opened, select the file, give the module a name, assign it a group (for sorting), comma separated tags (see "Using LLMs"), choose a color, because colors are great, and optionally set the clip on looping and set delay or volume ranges (or leave them at their default values). save the module, then click on the red X to close the popup window. 

Add more modules (or load existing ones with IMPORT), and save the scene (with SAVE ... obviously) so that you can reload it anytime you want.

Using AI:

You can use a local Large Language Model (commonly referred as 'AI') running via ollama (https://ollama.com) to generate soundscape automatically from prompts. The prerequisite is that you already created a sizeable library of sound modules and tagged them accordingly. You should add the tag 'Music' to the corresponding sound files, to give the LLM a chance to identify those correctly. You will also, obviously, need to download a model and have the ollama server running in the background.

Just enter the number of music files and the number of sound effects you want, and a description of the scene you need ("a lively medieval town. it's market day. Churchbells are ringing" for example ) and press enter in the prompt box. 

The LLM will now try to choose sounds that fits the scene.

Caveat and other things to be aware of :

• The LLMs can not perform miracles (yet?). The resulting soundscape if all your soundfiles are scifi spaceship sounds and you ask for a medieval fantasy world is bound to be disappointing, or at least "interesting".
• The context size (basically the short term memory of the LLM) is really important, especially if your library is vast. You can create alternate copies of models you already downloaded with higher context by using the Ollama command line chat with the model you want and typing :
  /set parameter num_ctx {the context you want in token}
  /save {a new name for the model} 
• I tried a few models, from 3b to 30b. Those which performed the best were Qwen3 and Qwen3-coder models. Other models (llama3.2, granite, gemma, phi, ... ) worked but had more trouble evaluating the prompt in the context of a soundscape ... but experiment yourself. Maybe I was just unlucky.
• Sometimes (especially with smaller models) things go wrong and the scene is full of invalid modules (or even no modules at all). Just send the prompt again to retry.

I hope you enjoy it and find it useful. I know I do, and that's probably the main point :)

Cheers,

Sammy