Setting maximum voices. about tinysoundfont HOT 6 CLOSED

So I've done a bit of digging....

I've duplicated my Bela code into a windows console application - this seems to run without any audio noise / trashing during render however many notes are played.

I'm guessing from this that my Bela is running out of CPU headroom (despite the CPU monitor never going over 50%). I've implemented some basic note stealing and limited Bela to 32 notes - I'm using the oldest note - and it works pretty well.

More when I've tidied the code a little.y

from tinysoundfont.

On what sort of device is the meter not going over 50%? On some platforms, the CPU usage percentage is across the whole CPU, and for a dual core CPU, 50% would be 100% of a single core. I think Linux and macOS both register percentage of individual CPU cores, rather than percentage of the entire CPU set available.

from tinysoundfont.

@adrianmcroft Do you know if the CPU in your Bela board has multiple cores? At least on Windows a reported 50% usage on a dual core CPU would mean one thread running at 100%. TSF itself runs only single threaded.

So yeah the current implementation doesn't limit max voices. In the function tsf_note_on on line 1251 it decides that it ran out of room for voices and allocates space for 4 more. You could fairly easy check if your maximum is reached and then just return; right there before the increase of voiceNum.
Adding something like a tsf_set_max_voices(...) function with a default of 0 (unlimited) voices would be a good idea I think.

from tinysoundfont.

Thanks Christopher and Bernhard, I'm still crossing over the stream to Linux and C++ from Windows and C# so thanks for any support.

So I'm running on a Beaglebone Black with a Bela Cape (BBB) which has a 1Ghz single Arm chip (http://bela.io/). Bela has very low latency which is why I've chosen it to play with.

So I've implemented some basic note stealing. As you say, before allocating new voices in [TSF_NOTE_ON] I'm counting playing voices and if its above my chosen maximum I steal. I'm using the oldest note - I've added a cycle counter INT to TSF_VOICE which is incremented during render and the biggest is the oldest. It's very rough but I'm just trying it as a POC. It actually works very well indeed.

Bela requires a funky cable to debug on the board and until them I'm stuck in suck and see land.

from tinysoundfont.

Ohh, yeah stealing old voices instead of just not playing new voices is a great idea.
You don't even need to count voices, you can just do the voice stealing inside the if (!voice) block as that means that all allocated voices are active. Also to find the oldest voice just check which voice has the oldest playIndex value. This value gets increased by one for every new voice.

from tinysoundfont.

Crumbs, that's a better tactic. I'll review the changes - I can now rip out half of my lines at least :)

from tinysoundfont.