I2S samples sometimes playing distorted?

[Lana-chan]

I'm not sure where or how to ask for help with this, so please let me know if more information is needed. I'm writing a sequencer-style program for the M5 Cardputer (ESP32-S3 based, 8MB Flash no PSRAM) and for that I've been developing a sound library which queues the playback of samples in memory with software mixing. Somehow this works fine on some programs and not in others, using the same MP firmware and same sound library.

I'm currently testing with the nightly build MicroPython v1.23.0-preview.176.g90e517862 on 2024-03-04 (This issue also appears in 1.21.0 and 1.22.0). In one simple program where all I do is load the samples and play a simple sequence of them, everything sounds fine. In another, more elaborate program which loads display drivers as well, suddenly all samples play as if they were volume boosted by a factor of 2. There is no cuts or dropouts in the sound, and if I set the volume (which my library calculates by bit shifting) lower, the sample appears to play fine (with a loss of bit resolution). Swapping between the two programs on the fly (running them as live scripts from something like Thonny) maintains this behaviour: one program plays fine, other seems overdriven and distorted, even though the library calls and samples are identical in both.

Even more bizarrely, sometimes this issue seems to go away if I completely clear my ESP by overwriting the MicroPython firmware and uploading files from scratch, without changing them. I've tried to trace down this fault to anything deterministic and I simply can't. Does anyone have any ideas?

[Lana-chan]

ha, what bad timing.

Just saw this pull request which replaces the I2S driver, and I was on a nightly just before that. Updated to v1.23.0-preview.203.gd712feb68 on 2024-03-09 and it seems to be working better than ever now, wasn't my code at all! In fact, this also fixed the issues I found empirically that only the right channel would play on the mono I2S device (NS4168) that the cardputer uses, forcing me to use I2S.STEREO and fill only the odd words in the buffer. Now mono works and I don't need to do any trickery!