All sounds play as noise exept saw and sine test on ESP32-A1S Ai audio kit V2.2 A436

[Magontek]

All notes the module play sound like noise.
I was able to play the saw test and sine test, but they sound strange—distorted, noisy, or crackly—though they do produce sound: saw example

For testing I'm using:
#define MIDI_DUMP_SERIAL2_TO_SERIAL
#define MIDI_RECV_FROM_SERIAL
#define MIDI_MONITOR_ENABLED

The MIDI command I send via serial is 0x903d7f, and the response I get is:

90 3d 7f<<\n> 0 | 90 3d 7f | Note On | Channel 0 | Pitch 61 | Velocity 127<\n>

Which seems correct.

But the resulting sound is like this, and clearly not right,

Things i've already try:

• Loading WAV files as full sample sets, as individual notes, and from SF2 soundfonts. Not sure if it matters, but all samples are loaded from FS_ID_SD_MMC.
• Testing all possible I2C and I2S configurations. The only one that works is: I2C 2 with I2S 1
  defined as:
  #define ES8388_CFG_I2C 2
#define ES8388_PIN_SDA 33
#define ES8388_PIN_SCL 32

#define ES8388_CFG_I2S 1
#define ES8388_PIN_DOUT 35
#define ES8388_PIN_DIN 25
#define ES8388_PIN_LRCK 26
#define ES8388_PIN_SCLK 27
#define ES8388_PIN_MCLK 0

• Lowering output volume using ES8388_SetDACVOL(0.4);. This became necessary, otherwise the saw wave is overly saturated.
• Testing with headphones. They work, but the sound remains the same.
• Loading other examples as ml_piano_example and ml_basic_sinth_example which sound like this

I’ve tried to review the I2S code, but haven’t found anything that could clearly explain this behavior.

Could it be a hardware issue with my board? I really hope not—I can't afford to replace it right now.

Any thoughts you might have, or things I could try, would be greatly appreciated.

I’m determined to get this module working, but I keep running into walls.

[marcel-licence]

I could only hear mouse clicks in the audio. Seems all three links referring to the same audio file.
I am not sure if there was a closed filter in the default setting of the project. Did you try out the piano example?
The note on message looks correct to me so far.

[Magontek]

I sorted the recordings I had made of the sounds. I don't know what happened with the GitHub attachment before.
I did try the piano example and I get this sound from Note 61
https://github.com/user-attachments/assets/a445bd50-f3d2-44a5-9b53-ecaeef53e9cd

[Magontek]

Update on my testing:

The board is definitely working properly. I tested it using this example from a different library, and it worked fine.

I’ve been trying to get a clean sawtooth output using ml_piano_example, but without success. I attempted to trace the audio data from the source to the driver at various points to identify where the signal degrades. So far, I can confirm that the data remains consistent between steps and still resembles valid audio.

To inspect the data, I used a very crude method—adding Serial.print statements in different parts of the code. For example, I inserted:

Serial.printf("%d, %d;", n, sampleDataU[n].ch[1]);

in i2s_interface.h, inside the function:

bool i2s_write_stereo_samples_buff(const float *fl_sample, const float *fr_sample, const int buffLen)

Obviously, the audio output becomes messy during this test due to the delays caused by serial printing.

When using the #define NOTE_ON_AFTER_SETUP option, I recorded this audio. It sounds clean, but it doesn't resemble what I hear from the speakers.

Could you please confirm whether this recording sounds anything like note 60 triggered by this line?

rhodes->NoteOn(ch, note + 12, vel);

This would help me narrow down where the issue is occurring.

[marcel-licence]

Hm that sounds strange. Seems that there is some problem.
I was working with the E-Piano today, which worked on a normal ESP32. https://youtu.be/c0i5jDp6_rQ
Do you have other changes than in ES8388.h ?