shortcomings in audsrv and upsampling

[irixxxx]

While working on PicoDrive I found the audsrv implementation a bit lacking:

• If the audsrv backend on IOP runs low on samples it will replay the buffer over and over, right down to telling you about the buffer being miraculously filled again. New samples then don't have space in it, and the reverberating audio you get remembers me of old GSM mobiles in bad reception conditions. It simply can't discern "ringbuffer full" from "ringbuffer empty"... so it wraps around and around.

• it sets the ringbuffer to half full on set_format without clearing it. Hence, there's always some old audio from the buffer played out first.

• stopping will keep all unplayed samples and there is no way to clear the audio ringbuffer.

[AKuHAK]

nice, I also noticed something similar, do you know the solution?

[irixxxx]

If I had one I would've made a PR ;-)

Most of it should be easy to fix, but it potentially breaks stuff relying on those bugs. I guess it's up to the maintainer (if there is one at all).

As for the bugs:

• for the ringbuffer, either add an "empty" flag (e.g. a count of samples in the ringbuffer), or make sure it's only filled up to (size-1)
• on set_format, I'd just assume the buffer to be cleared
• likewise on stopping, or if that's not desired, add a clear buffer function

[irixxxx]

The upsamplers used to upsample to 48 KHz are also flawed:

• the counts returned by the upsamplers are off in at least one case, e.g. for up_11025_16_mono it returns 234, but 2x512x11025/48000 is 235.
• there is no fractional part to it, which leads to the frequency being slightly off by up to 1 input sample for 512 output samples (up to 0.5% depending on the input sample rate). Nonetheless, if the sample producer delivers the exact amount, there needs to be some glue logic for resampling (at least by throwing away some samples).

[irixxxx]

That upsampling stuff really bothers me. I don't think using lookup tables is a good solution to handle this.

I had a similar problem with resampling the 44.1 KHz CD audio tracks to the target sample rate. I'd suggest whoever is going to address this might want to look up my solution in the function mix_16h_to_32_resample_stereo in pico/sound/mix.c. It's also relatively independent of the source and target sample rates, so the restriction to certain rates might potentially also be lifted.