24 bit it is time! :)

[Marcin-Fenger]

Hi Wolle

I have written and successfully tested 24‑bit conversion with noise shaping on the same reinterpreted buffer. Perhaps you would like to test it as well. Now the doors are open to supporting 24‑bit FLAC :) What do you think about it?

Rebuilded playChunk: https://pastebin.com/FHtvNwXm

Now it is working with 48kHz conversion disabled. You should reconfigure I2S for 24bit as well and set output_24bit true. 32 bit should working without problems.

Have fun :)

[schreibfaul1]

Great, I'll take a look at that soon.

[dzikakuna]

Marcin, can shorty describe how to implement it for testing ? Sounds like opening completely new quality for this library.
What is the maximum bitrate possible to achieve on 24 bits ?

[schreibfaul1]

In the meantime, I have expanded the OutBuff from int16_t to int32_t. This is necessary for true 24-bit playback (int24_t does not exist).
This allows WAV files with 24 or 32 bits per sample to be played back. An S3 or P4 is required for this; the old ESP32 is too slow.
Later, I will adapt the FLAC decoder for 24-bit.

[Marcin-Fenger]

Hi Wolle. At this moment I am testing on old ESP32 with 48k disabled, FIR disabled and vu disabled. Everything works as expected. Changing to int32 is good direction but We should remember that native 16 bit needs int16_t. Support 24 bit flac with 16 bit output should be downconverted. I think 24bit FLAC should be played in 24/32 bit setup only. Have You tested my 24 bit solution yet?

[schreibfaul1]

Thank you for your feedback. The decoders (mp3, aac, opus, vorbis) do indeed require the int16 format.
This trick int16_t* out16 = reinterpret_cast<int16_t*>(outbuf);
ensures that the decoders do not notice that outbuf is int32_t*

I will be upgrading the FLAC decoder to 24-bit next year.
You can already test 24-bit WAV files. Nothing is decoded with WAV, it's just little endian.

[dzikakuna]

Wolle, thanks for today commit, 24bits FLAC stations starts to work, not all but most of them. Some issue is now with TTS. Sound of TTS (used by me for voice clock) is like mix TTS with noise.