How to make Miniaudio play multiple sounds from the same data source more than once?

[GasimGasimzada]

I am using miniaudio High level API with resource management disabled:

#define MA_NO_RESOURCE_MANAGER
#include <miniaudio/miniaudio.h>

I have a custom resource registry that loads audio files and stores the ma_decoders:

ma_decoder *decoder = new ma_decoder;
auto res = ma_decoder_init_file(filePath.string().c_str(), nullptr, decoder);

if (res != MA_SUCCESS) {
  return Result::Error(...);
}

mRegistry.getAudios().addAsset({ decoder });

In my audio engine, I create sounds from the same decoder and play them. Once the sounds are done playing, they get uninitialized and deleted. Here is how I create, play, and uninitialize sounds.

void *playSound(void *dataSource) {
  auto *sound = new ma_sound;

  auto initRes =
      ma_sound_init_from_data_source(&mEngine, dataSource, 0, nullptr, sound);
  LIQUID_ASSERT(initRes == MA_SUCCESS, "Cannot init sound");

  // play the sound
  auto startRes = ma_sound_start(sound);

  return sound;
}

void destroySound(void *sound) {
  ma_sound_uninit(static_cast<ma_sound *>(sound));
  delete sound;
}

This particular setup works for a single sound. If I have two separate assets (aka two decoders), I am able to play sound from each decoder once. However, if I try to create multiple sounds using the same decoder as the data source, the sound only gets played once.

While debugging this issue I noticed that, on second play, the sound was initialized successfully, start function returned MA_SUCCESS but ma_sound_is_playing immediately returned false:

// new data source -- sound plays
void *sound = playSound(decoderBark);
ma_sound_is_playing(sound); // true

// 5 seconds later
{
  // new data source -- sound plays
  void *sound = playSound(decoderMeow);
  ma_sound_is_playing(sound); // true
}

{
  // data source already used -- sound does not play
  void *sound = playSound(decoderBark);
  ma_sound_is_playing(sound); // false
}

// 5 seconds later
{
  // data source already used -- sound does not play
  void *sound = playSound(decoderMeow);
  ma_sound_is_playing(sound); // false
}

{
  // data source already used -- sound does not play
  void *sound = playSound(decoderBark);
  ma_sound_is_playing(sound); // false
}

// ... none of the already loaded sounds ever play

[mackron]

That's not how miniaudio works. You create a separate data source for each sound you want to play. The resource manager is used for managing caching and loading sound data only once.

[GasimGasimzada]

Since I am not using the resource manager, is it possible for me to derive one data source from another one? I have been reading docs about the AudioBuffer and there is this snippet:

ma_audio_buffer_config config = ma_audio_buffer_config_init(
    format,
    channels,
    sizeInFrames,
    pExistingData, // <-- Can this be the data stored in decoder
    &allocationCallbacks);

I would like to know if I can create a decoder using ma_decoder_init_file and store it as my "original asset". Then, any time I need to play a sound, I want to create an audio buffer (or another decoder) by just using the same buffer that is stored in the original asset. Is that possible?

[GasimGasimzada]

I think I figured out how I am going to do this. I am going to load the raw binary data into memory using my own methods (e.g std::ifstream), then call ma_decoder_init_memory with config that identifies what decoder to use (WAV, FLAC, or MP3).

[mackron]

There's a few options. It'll involve you having to do some memory management yourself.

You can load the file data of your sound into a buffer and use ma_decoder_init_memory(). That will not make a copy of the data buffer so you won't have any significant duplication of data.

The above option will result in each sound object having to spend time doing the actual decoding. If you wanted to avoid that, you could pre-decode the audio data and then use an ma_audio_buffer for each sound. The downside to this is that it'll use more memory.

[GasimGasimzada]

The downside to this is that it'll use more memory.

Could you please explain why it would use more memory? Does audio buffer duplicate allocate the provided data and copy it? I think this is what I need in the long run: Precode the data before hand and use audio buffer to store that memory for each.

[mackron]

Because decoding will decompress the audio data. Like an MP3 or Ogg file, for example. When using ma_decoder_init_memory() you can just pass in the raw audio data which will be compressed, but when you decode that data it'll be in uncompressed PCM format.