Examples or Docs re. the Callback on ma_engine

[kunle-e]

In the change log for v0.11.18 the following is mentioned: Add support for setting a callback on an ma_engine object that get's fired after it processes a chunk of audio. This allows applications to do things such as apply a post-processing effect or output the audio to a file.

Where can one learn how to work with this callback on ma_engine--is there any documentation on it yet?

[mackron]

There's no documentation, but it's easy to set up. You just do it via the engine's config:

void do_processing(void* pUserData, float* pFramesOut, ma_uint64 frameCount)
{
    // Do something with `pFramesOut`. Will contain valid audio data on input. Modify it however you like.
}

ma_engine_config engineConfig = ma_engine_config_init();
engineConfig.onProcess = do_processing;
engineConfig.pProcessUserData = NULL; // Set this to a pointer of your choice (or NULL). It will be passed into the `pUserData` parameter of your callback.

ma_engine_init(...)

[kunle-e]

Hi @mackron, I was able to take a look at this today. It seems like the do_processing() function is only called when ma_engine_read_pcm_frames(&g_engine, pSomBuffer, framesChunk, &pFramesRead); is called, which can happen in the data_callback() or outside the data_callback().

My concept of this might be way off, but I'm trying to understand how useful do_processing() would be for outputting the audio to a file since most likely one would not want to output to file when calling ma_engine_read_pcm_frames() from the data_callback() because it would take however long the audio is in time to write the file, but I would likely want to write the file it in a loop outside the data_callback() where ma_engine_read_pcm_frames() is used to populate a buffer for writing, which means that I would write the file in this same loop negating the need for the do_processing() callback.

Initially I thought one advantage of do_processing() is that you don't have to create a buffer, but since I would need to call ma_engine_read_pcm_frames() outside of the data_callback(), then I would need to allocate a buffer anyway. Perhaps I can call ma_engine_read_pcm_frames() without populating a buffer, which would firedata_callback()? How should I be looking this?

[mackron]

Yes, that is correct - it is fired from ma_engine_read_pcm_frames(), which will be called from the data callback on the audio thread. Calling straight into a file write operation from there is indeed not ideal. However, no matter what technique you use, you still need to get that data somehow so at some point you'll need to call ma_engine_read_pcm_frames().

This to me sounds like a classic use case for a lock-free ring buffer. The data callback (audio thread) would be the producer which writes into the buffer, and then another thread (the consumer) reads from the buffer and outputs the data to a file. You would just need to make sure the ring buffer is big enough to be able to cache extra data from the producer to deal with delays by the consumer thread. But nothing is really perfect with this - your consumer might never be able to keep up with the producer which will result in glitching in the recorded output. It's all about how advanced you want to go with it and the trade-offs you're willing to accept.

[kunle-e]

Thanks, makes sense.