RP2040: audiobusio.I2SOut: on playing wav from file: i2s.playing False too early and random

[dglaude]

Testing I2SOut on pico + "pico base" with 6.2.0-beta.2-154-gd4bf0d5e2

("pico base" use PCM5100A DAC for line out audio over I2S, just like in Raspberry Pi reverence design)

Playing "StreetChicken.wav" from flash on I2S, the file and code inspiration come from the learn guide: https://learn.adafruit.com/adafruit-i2s-stereo-decoder-uda1334a/circuitpython-wiring-test

The sound is metalic (randomly, sometime it is good) and i2s.playing return False "randomly" long before the end of the sound file.

Code:

import board
import audiocore
import audiobusio
import time

print("Playing 10 times StreetChicken.wav")

f = open("StreetChicken.wav", "rb")
wav = audiocore.WaveFile(f)
i2s = audiobusio.I2SOut(board.GP27, board.GP28, board.GP26)

for i in range(10):
    start=time.monotonic()
    i2s.play(wav)
    while i2s.playing:
        pass
    stop = time.monotonic()
    print("Total lenght = ", stop, "-", start, "=", stop-start)

Expected result:

Perfect sound, the wave file should play 10 times and last more or less exactly the same length.

Effective result:

The sound is unperfect (metalic scratch) and the timing is random

Playing 10 times StreetChicken.wav
Total lenght = 972.984 - 972.949 = 0.0349121
Total lenght = 973.148 - 972.985 = 0.163086
Total lenght = 974.415 - 973.149 = 1.26587
Total lenght = 974.985 - 974.416 = 0.568848
Total lenght = 976.125 - 974.987 = 1.13818
Total lenght = 980.144 - 976.126 = 4.01782
Total lenght = 981.144 - 980.145 = 0.998779
Total lenght = 981.947 - 981.145 = 0.802002
Total lenght = 983.191 - 981.948 = 1.24292
Total lenght = 983.877 - 983.192 = 0.685059

[dglaude]

Replacing

while i2s.playing:
    pass

by `time.sleep(15)`

Make a much better sound (except the first play) and let you hear the full sound.

> Playing 10 times StreetChicken.wav
> Total lenght =  258.805 - 243.804 = 15.001
> Total lenght =  273.807 - 258.806 = 15.0011
> Total lenght =  288.809 - 273.808 = 15.001
> Total lenght =  303.811 - 288.81 = 15.001
> Total lenght =  318.813 - 303.812 = 15.001
> Total lenght =  333.815 - 318.814 = 15.001
> Total lenght =  348.817 - 333.816 = 15.001
> Total lenght =  363.819 - 348.818 = 15.001
> Total lenght =  378.821 - 363.82 = 15.001
> Total lenght =  393.823 - 378.822 = 15.0011

[dglaude]

This give much better result (so replacing pass by a short sleep):

    while i2s.playing:
        time.sleep(0.01)

Still the first play can be too short and/or metallic.

[recantha]

Just wanted to confirm that this happens as David says. It's odd. It's the kind of thing you think "should work" but somehow it doesn't quite. Impressive stuff, though, what goes on under the hood.

[tannewt]

I think next step to debugging this is to capture the I2S output stream and then compare it to the original file. Doing that would allow one to characterize how the playback is failing. (Repeated regions of the source for example.)

[bhilimon]

I've been working on getting audio (wav) working on my Feather RP2040 and I'm also having these similar issues but wildly inconsistent behavior. Mine is coupled with the Adafruit I2S mono amp (https://www.adafruit.com/product/3006). Using the same StreetChicken.wav test file. I'm not sure how far long I2S audio is on this new board but I figured I'd throw in what I'm experiencing if I can help at all.

Is there a preference to what pins we should be using for I2S audio? My setup is as follows:

audio = audiobusio.I2SOut(board.D11, board.D12, board.D25)

It does play audio but I have the following inconsistent/random issues:

• Most of the time there's some static or other noise coming through along with the audio. Pretty much guaranteed on the first play, and maybe 50/50 on future plays. Sometimes the additional noise will cut in or out mid-play. Only sometimes the audio comes through clear for it's entirety. It doesn't seem to be a gain/volume issue. I've been playing around modifying the gain on the amp board but seem to have the same results no matter what. Also swapping between 2 different speakers. I've played with 3.3v and 5v to the amp, that also doesn't seem to matter either.

• Sometimes the playing() function gets "stuck" and always reports True even after the audio file is done playing.

• Sometimes the audio cuts off early. The playing() function seems to change back to False when this happens.

I'll continue trying things with my setup and report back if I make any progress.

[tannewt]

The pins shouldn't matter because it's all PIO under the hood. It's likely a timing hiccup with the DMA and sample conversion code.

[bhilimon]

I forgot to reply back with a few notes after playing around yesterday.

• I was trying to play audio while also running some NeoPixel animations. When both were playing the audio distortion would randomly come in and out. Disabling or pausing the NeoPixel animations while audio is playing seems to have completely cleared up the audio distortion except for the initial audio file (next bullet). I don't know if this is a CPU/memory thing or if I just can't do audio and animations at the same time. Either way, I can work around that.
• The first audio file that is played is still always distorted. It doesn't matter how long it is (.wav). The second time an audio file plays it comes through fine, even if it's a different file. I was thinking it might have been something with loading the file from flash, but doesn't seem to be as subsequent files play fine.
• As a test I moved to a PIO "group" that is not shared with other used pins, that didn't seem to help things.
• The audio.playing() function still appears bugged. I worked around that one for now too.
• I also swapped in my analog amp and switched to audiopwmio instead of audiobusio (i2s) and the distortion on the first file problem followed it. I don't think I tested the playing() function.

Hope it helps.

[rsbohn12]

Similar results trying to play a RawSample on the Feather RP2040 with MAX98537.

Adafruit CircuitPython 7.0.0-alpha.2-646-g7df5d74d9 on 2021-05-22; Adafruit Feather RP2040 with rp2040

dac = audiobusio.I2SOut(bit_clock=board.D5, word_select=board.D6, data=board.D9)
clip = make_squarewave(440*Hz, duty_cycle=0.5)
dac.play(clip)
time.sleep(0.5)
dac.stop()

I hear a click at dac.stop() but not audible tone is produced. I measured a 320 kHz signal on the MAX98537 terminal block.

When I connect the same breakout board to the SAMD21 QT Py it does produce audible tones.
I also tried Adafruit CircuitPython 6.2.0 on 2021-04-05; Adafruit Feather RP2040 with rp2040 -- still no sound.