adlmidi

adlmidi copied to clipboard

on archlinux, with an ordinary alsa / pulseaudio configuration, adlmidi plays back audio at a barely audible volume. Running with -w produces a .wav file with a very low volume as well.

That's because the gaining is needed (in the code I mean) to resolve this problem. Also, you could disable the reverbation that also reduces the volume.

Thanks. Unfortunately, running with -nr gives a segfault :

/usr/include/c++/13.2.1/string_view:258: constexpr const std::basic_string_view<_CharT, _Traits>::value_type& std::basic_string_view<_CharT, _Traits>::operator[](size_type) const [with _CharT = char; _Traits = std::char_traits<char>; const_reference = consAborted (core dumped)

This, seems, an another and unrelated bug...

The default settings in reverb reduce the gain in order to avoid clipping problems. This could be helped by doing some DC offset removal before running reverb (and would be a good idea in general).

The default settings in reverb reduce the gain in order to avoid clipping problems.

Thanks for the explanation. I'm surprised it needs this much attenuation, it sounds about 30dB too quiet...

I agree, but DC offset can make all the difference. If you don't know what I mean, consider this. If the sound alternates only between values +2 and –1, its amplitude would only be 3. Now, if you multiply that with just 1.0001, you get an alternation between +2 and –1. If you express this in 16 bits, you still get –2 and +1, an amplitude of 3. However, if the sound alternates only between values -32765 and -32768, yes, its amplitude would still only be 3. However, if you multiply that with just 1.0001 (same as before), you get an alternation between –32768 and –32771. If you express this in 16 bits, you get values –32768 and +32764, an amplitude of 65514, which is many orders of magnitudes larger than before. This is why the DC offset can make all the difference.

Another solution is just an output as Float32. Unlike integer formats, float allows to clip and don't distort the result.

Yes, float32 format is a way of saying it’s someone else’s problem now. (I don’t know why ADLMIDI doesn’t use float32 for WAVs.) ADLMIDI actually does make an attempt at negating the DC offset, but it is not perfect.