view mixer/tutorial.txt @ 552:2e8907ff98e9

Replaced references to COPYING with references to LICENSE.txt ...
author Ryan C. Gordon <icculus@icculus.org>
date Tue, 09 Sep 2008 16:28:11 -0400
parents 3a3807dcf57f
children
line wrap: on
line source


SDL_sound version 2.0: Your Mixer, And Welcome To It.


SDL_sound v2's major addition is a software mixer. There are some other new
features, but this is the Big New Thing.


Some notable features of this mixer:
 - No clamping at mix time. Most SDL-based mixing is done via SDL_MixAudio(),
   which mixes two samples together, clamping to 16-bits, and then mixes the
   next sample into this already-clamped buffer. This can lead to audio
   distortion. SDL_sound mixes all samples into a 32-bit floating point buffer
   before passing it to the audio device, which means no unnecessary clamping.
   It also means it can be more optimized for MacOS X's CoreAudio and other
   next-gen audio subsystems, which require you to feed it float32 data.

 - Optimized mixing: MMX, SSE, 3DNow!, and Altivec are used internally where
   appropriate. (er...they WILL be, at least!)

 - Multiple "music" files. SDL_mixer is notorious for making a distinction
   between "sound" and "music" files (that is, things that can be trivially
   decoded to a waveform without bloating the memory footprint vs. things that
   can't), and only allows mixing of one music file at a time. SDL_sound
   doesn't bother with this distinction, which means you are free to mix any
   combination of audio formats at the same time.

 - No "channels". If you want to mix 1,000 audio files and have the hardware
   to cover it, you can. You don't have to manage playback channels. There
   isn't a seperate "music" channel, since "music" isn't treated differently
   from any other audio.

 - Lots of formats. SDL_sound already decodes a huge number of audio formats.
   As the mixer is layered on top of this, all of the format support comes
   for free.

 - Can handle non-power-of-two resampling. If your samples are at 8000Hz
   and the audio hardware is at 11000Hz, this doesn't cause output problems.

 - Control over buffering and mixing. SDL_sound already lets you control how
   much audio is prebuffered and predecoded; this carries over to the mixer,
   so you can customize how your resources are being used on the fly.

 - Flexible enough for those that need to micromanage mixing, but dirt simple
   to for those that just want to make some noise.

 - It can be compiled out if you just want the 1.0 API for decoding formats.
   (initializing the mixer subsystem in this case will fail, but for binary
    compatibility, the entry points will still exist).

A brief tutorial follows.


Example #1: Play a sound and get the heck out of there.

#include "SDL_sound.h"

int main(int argc, char **argv)
{
    Sound_MixInit(NULL);  // start the mixer; don't care what format.
    Sound_Sample *hello = Sound_NewSampleFromFile("hello.wav", NULL, 10240);
    Sound_MixPlay(hello);
    while (Sound_MixPlaying(hello))
        SDL_Delay(100);  // wait around; mixing is in a seperate thread!
    Sound_FreeSample(hello);
    Sound_MixDeinit();
    return(0);
}

Every tutorial needs a "Hello World" example.
That will play hello.wav, wait for it to finish, and terminate the program.
But that's not really mixing! To qualify, you'd need to play two sounds at
once. So let's do that:


Example #2: Mixing two sounds.

#include "SDL_sound.h"

int main(int argc, char **argv)
{
    Sound_MixInit(NULL);  // start the mixer; don't care what format.
    Sound_Sample *hello = Sound_NewSampleFromFile("hello.wav", NULL, 10240);
    Sound_Sample *music = Sound_NewSampleFromFile("icculus.ogg", NULL, 10240);
    Sound_MixPlay(music);
    while (Sound_MixPlaying(music))
    {
        if (!Sound_MixPlaying(hello))
        {
            Sound_Rewind(hello);
            Sound_MixPlay(hello);
        }
        SDL_Delay(100);  // wait around.
    }
    Sound_FreeSample(music);
    Sound_FreeSample(hello);  // will stop if it happens to still be playing.
    Sound_MixDeinit();
    return(0);
}

Same deal, but we play some music ("Icculus in San Jose" from the talented
Emmett Plant, in this case). We also load our "hello" sound from the previous
example. While the music is playing, we check if "hello" is playing, and if
not, we set it up to play again. Note that the two sounds are playing at the
same time, mixed together. Cool, huh?

You might notice that we called Sound_Rewind() on the hello sample. This isn't
part of the mixer itself, and is a function from SDL_sound v1, before there
was a mixer at all. This illustrates that you can use the usual SDL_sound
methods to manipulate a sample in the mixer, including seeking and predecoding.
These are safe operations even while the sample is playing.

That's about all you need to know to effectively use the mixer. Everything
after that is extra credit.


Extra credit #1: Mixer Attributes.

An API's got to know its limitations. SDL_sound isn't meant to be a robust 3D
spatialization library. For that, one should look to the excellent OpenAL API
at http://www.openal.org/. Still, for many reasons, OpenAL might not be a good
fit: it doesn't support many audio formats (and all formats except uncompressed
integer PCM are optional extensions to the API), it is less likely to
support your platform and audio output target than SDL, and it is more
complicated to feed it streamed audio. While not as robust as AL's feature
set, SDL_sound v2 provides a handful of useful attributes you can set on a
sample to alter its playback.


Basic Attribute #1: Looping.

Checking a sample's playing state in a loop just so you know when to restart
it has two problems: first, it's a pain in the butt, and second, there may
be a gap in the audio between when the sound starts and when you're able to
restart it. To remedy this, SDL_sound lets you flag a sample as "looping" so
you don't have to micromanage it. It will continue to rewind and play until
you explicitly stop it. Let's take our last example and do this right:


Example #3: Mixing two sounds with better looping.

#include "SDL_sound.h"

int main(int argc, char **argv)
{
    Sound_MixInit(NULL);  // start the mixer; don't care what format.
    Sound_Sample *hello = Sound_NewSampleFromFile("hello.wav", NULL, 10240);
    Sound_Sample *music = Sound_NewSampleFromFile("icculus.ogg", NULL, 10240);
    Sound_SetAttribute(hello, SOUND_ATTR_LOOPING, 1);  // turn on looping.
    Sound_MixPlay(music);
    Sound_MixPlay(hello);
    while (Sound_MixPlaying(music))
        SDL_Delay(100);  // wait around.
    Sound_FreeSample(music);
    Sound_FreeSample(hello);  // will stop now.
    Sound_MixDeinit();
    return(0);
}

...it's that easy.


Basic attribute #2: Fire and forget

You'll notice in previous examples that we are taking the pains to explicitly
free the resources associated with a sample via the Sound_FreeSample() call.
In a small program like this, it's easy to be tidy and sweep up after one
or two hardcoded sounds, but when you are managing a lot of different sounds,
or a lot of copies of the same sound, this can become tedious. Case in point:
laser beams.

Let's say you've got a space fighter game, with a bunch of ships flying
around and shooting at each other. Every time they fire a laser, do you really
want to take the effort to decide when it is done and clean it up? You want
to, quite literally in this case, "fire and forget" the sound...that is, you
want the mixer to playback the audio and then clean it up without further
action or intervention from you.

So let's take our previous example and adjust it to clean up after us.


Example #4: Fire and forget playback.

#include "SDL_sound.h"

int main(int argc, char **argv)
{
    Sound_MixInit(NULL);  // start the mixer; don't care what format.
    Sound_Sample *hello = Sound_NewSampleFromFile("hello.wav", NULL, 10240);
    Sound_Sample *music = Sound_NewSampleFromFile("icculus.ogg", NULL, 10240);
    Sound_SetAttribute(hello, SOUND_ATTR_FIREANDFORGET, 1);
    Sound_SetAttribute(music, SOUND_ATTR_FIREANDFORGET, 1);
    Sound_MixPlay(music);  // play once, then call Sound_FreeSample() for you.
    Sound_MixPlay(hello);  // play once, then call Sound_FreeSample() for you.
    while (Sound_MixPlayingCount() > 0)
        SDL_Delay(100);  // wait around.

    // Don't need Sound_FreeSample() here anymore!

    Sound_MixDeinit();
    return(0);
}

So everything was deallocated automatically, and your mother didn't even have
to come along and tell you to clean up this pig sty! She is very proud of you
right now, I assure you.

You'll note that we call Sound_MixPlayingCount() to see if the music finished.
You have to do this because the "music" sample is invalid once it gets pushed
through Sound_FreeSample(), which will happen as soon as the mixer is done
with it. To avoid touching deallocated memory, we just ask the mixer if
anything is still playing.

Also, common sense dictates that looping sounds never get to the "forget"
part of "fire and forget", since they don't stop playing. You can either
manually halt them or turn off the looping, though, and then they'll clean
themselves up.


Basic attribute #3: Per-channel Gain.

If you can tweak the volume of the left or right channel on a sample, you can
accomplish (or at least fake) a surprising number of simple sound effects.
Therefore the mixer allows you to do just this, and then builds a few features
on top of this magic.

This is accomplished by tweaking the "gain" of a given channel. "Gain" is just
a fancy way of saying "volume". You specify it as a floating point number,
usually in the range of 0.0f to 2.0f. If you set the gain to 0.0f, it results
in silence, and 1.0f results in no change at all. 0.5f halves the volume and
2.0f doubles it. As you might have guessed, the sample gets multiplied by
this value.

SDL_sound's mixer lets you tweak each channel in a sample individually. Right
now we're limited to mono (one channel) and stereo (two channel) sounds, but
this will probably be enhanced at some point. It's worth noting that this
refers not to the sample itself but to the speakers where they play. This
means you can set the left and right channels of a sample, even though the
sample itself only has one. Since a 2-speaker setup will promote a mono sound
to stereo (same waveform is fed to each speaker), you can tweak it to play at
different volumes in the left and right.


So to rehash our tired hello world example again...

Example #5: Per-channel gain.

#include "SDL_sound.h"

int main(int argc, char **argv)
{
    Sound_MixInit(NULL);  // start the mixer; don't care what format.
    Sound_Sample *hello = Sound_NewSampleFromFile("hello.wav", NULL, 10240);

    // Every channel's gain defaults to 1.0f, or no adjustment.

    Sound_SetAttribute(hello, SOUND_ATTRGAIN0, 0.0f);  // left chan==silence.
    Sound_MixPlay(hello);  // plays just right channel.
    while (Sound_MixPlaying(hello))
        SDL_Delay(100);  // wait around.

    Sound_SetAttribute(hello, SOUND_ATTRGAIN0, 1.0f);  // left chan==normal
    Sound_SetAttribute(hello, SOUND_ATTRGAIN1, 0.0f);  // right chan==silence
    Sound_MixPlay(hello);  // plays just left channel.
    while (Sound_MixPlaying(hello))
        SDL_Delay(100);  // wait around.

    Sound_FreeSample(hello);
    Sound_MixDeinit();
    return(0);
}


This played the hello sound twice, once in each speaker. Simple.

Well, almost simple. If you only have mono output (one speaker), then this
will play silence the first time (channel 0 set to silence), then the sound
at normal volume the second time (channel 0, the only speaker, set to normal).
In a 1-speaker setup, screwing with the second channel is ignored.

If this is going to be a pain for you to track yourself, you can use
Sound_MixInit() to set up a stereo environment and let it dither everything
down to one speaker behind the scenes if need be. Generally, this isn't a
huge concern, though.


Extra Credit #2: Fading

Sometimes you want to fade out (or fade in) a sound over time...this is
handy when ending a game level. It's a nicer effect to silence everything
over some small amount of time than to abruptly kill all the noise. This is
more pleasant for the end-user.

You could accomplish this by tweaking each channel of all your samples' gain
over time, but this is another one of those things that are annoying to
micromanage. The mixer has to constantly pay attention to these samples anyhow,
why should you do it, too?

SDL_sound gives you a means to instruct the mixer to take care of this.


Example #6: Fading a sound.

#include "SDL_sound.h"

int main(int argc, char **argv)
{
    Sound_MixInit(NULL);  // start the mixer; don't care what format.
    Sound_Sample *music = Sound_NewSampleFromFile("icculus.ogg", NULL, 10240);
    Sound_MixPlay(music);  // Start music playing.
    Sound_SetAttribute(music, SOUND_ATTR_FADEOUT, 10000);
    SDL_Delay(10000);
    Sound_SetAttribute(music, SOUND_ATTR_FADEIN, 10000);
    SDL_Delay(10000);
    Sound_FreeSample(music);
    Sound_MixDeinit();
    return(0);
}


So this starts our favorite song playing, and tells it to fade to silence
smoothly over 10000 milliseconds (that is, 10 seconds). Since we know how
long we want this to take, we lazily call SDL_Delay() to wait that long; the
mixer works in another thread, so we have the luxury of doing nothing here.
Then we fade it back in over another 10 seconds before exiting.

It's worth noting a few things here:
First, the FADEOUT attribute uses the same mechanism as SetGain() under the
hood when mixing, but the two attributes exist seperately: if you double the
gain (2.0f), the sound will drop in volume twice as much each time the fading
updates, but it's still going to go from twice-as-loud to silence in the
same amount of time (10000 milliseconds in this case).

When a sample is totally silent, either because it faded out or you set its
gain to 0.0f, it is still playing! If you were to turn the volume back up
30 seconds after the fade completes, you'd hear the sound as it would be at
that 30 second moment as if you hadn't silenced it at all. This has a few
important ramifications:
  1) It's still taking CPU time. Maybe not as much, since we can choose not
     to mix when the gain is 0.0f, but in order to keep the sound "playing"
     we might need to decode more of it, which means CPU time and memory
     usage and such. Best to halt a silent sound if you aren't going to need
     it.
  2) Sound_MixPlayingCount() might be > 0 even though you don't hear noise.
  3) A sound might not be where you left it. Keep better track of your things!

You might also notice that we called Sound_FreeSample() on a playing sample.
This is legal. If a sample is playing when you free it, the mixer knows to
halt it first.


Extra Credit #3: Halting

SDL_sound's mixer is a little different than most, in that there aren't
seperate playing states. "Halting" a mixing sample means you took it out of
the mixing list. "Playing" it means you put it back in, and it picks up the
mixing with the sample as it found it.

If you want something anologous to that "stop" button in WinAmp, you would
halt the sample and then call Sound_Rewind() on it. Next time you start it
playing, it'll be playing from the start of the sample. If you didn't call
Sound_Rewind() first, it'll be playing from where you halted it. That's more
like clicking WinAmp's "pause" button.

However, there are times when you want everything to stop at once. Just
looping over every sample and halting them isn't any good, since some might
play just a tiny bit longer...it's a lovely bug called a "race condition".

And, as I'm sure you've heard me say before, why do a lot of work to manage
stuff that the mixer has to manage itself anyhow? You should learn to
delegate more, you control freak.


Example #7: Halting.

#include "SDL_sound.h"

int main(int argc, char **argv)
{
    Sound_MixInit(NULL);  // start the mixer; don't care what format.
    Sound_Sample *hello = Sound_NewSampleFromFile("hello.wav", NULL, 10240);
    Sound_Sample *chatter = Sound_NewSampleFromFile("chatter.wav", NULL, 10240);
    Sound_Sample *music = Sound_NewSampleFromFile("icculus.ogg", NULL, 10240);
    Sound_MixPlay(music);

    Sound_MixPlay(hello);
    while (Sound_MixPlaying(hello)) SDL_Delay(100);  // wait around.
    Sound_MixPlay(hello);
    while (Sound_MixPlaying(hello)) SDL_Delay(100);  // wait around.
    Sound_MixPlay(hello);
    while (Sound_MixPlaying(hello)) SDL_Delay(100);  // wait around.

    Sound_MixHalt(music);  // halt the music.

    Sound_MixPlay(hello);
    while (Sound_MixPlaying(hello)) SDL_Delay(100);  // wait around.
    Sound_MixPlay(hello);
    while (Sound_MixPlaying(hello)) SDL_Delay(100);  // wait around.
    Sound_MixPlay(hello);
    while (Sound_MixPlaying(hello)) SDL_Delay(100);  // wait around.

    Sound_MixPlay(music);  // start the music where it left off.
    Sound_MixPlay(chatter);  // start the chatter.

    SDL_Delay(3000);  // let them play.
    Sound_MixHalt(NULL);  // halt _everything_ that's playing.
    SDL_Delay(3000);  // waste some time.
    Sound_MixPlay(music);  // start the music where it left off.
    Sound_MixPlay(chatter);  // start the chatter where it left off.
    SDL_Delay(3000);  // waste some more time.

    Sound_FreeSample(music);    // clean up and quit.
    Sound_FreeSample(chatter);
    Sound_FreeSample(hello);
    Sound_MixDeinit();
    return(0);
}


Ok, you following? That plays the music, plays "hello" three times while the
music is playing, halts the music and plays hello three times without anything
else, restarts the music where it left off mixed with "chatter" for three
seconds, stops everything (music and chatter in this case), waits three more
seconds of silence, restarts the music and chatter where it left off and
lets them play for three more seconds. Then it shuts it all down and goes
home.

Fun, huh?

There are some notable exceptions to this rule. When a sound gets to the end
of its mixing, it either halts or (if set looping) rewinds and starts playing
again without missing a beat. For these, you don't have to manually halt
(or manually restart, as it were).


You now have everything you need to make a game with robust audio. But for
some of you, that's not enough. You're never satisfied. You need the section
of this tutorial written for...


THE HARDCORE.

(These sections are brief and lack full examples. If you're hardcore, you
probably don't read wussy tutorials anyhow.)


Hardcore #1: low-overhead copying of a sample.

Let's say you've got a sound file that represents a laser blast. Everytime a
shot is fired in your game, you don't want to have the overhead of reloading
it from disk, decoding and mixing it on the fly!

Here are some tips for your efficiency:

    - Opt for an uncompressed format, such as a standard .WAV file or even
      raw PCM. For small, frequently used sounds, the bigger memory footprint
      is usually an acceptable tradeoff to constant redecoding. In some cases,
      we've found that compressing to, say, .ogg format actually makes the
      file bigger if it's a very short sound.

    - Put your sound into a memory block and point multiple memory RWOPS at
      it: one per playing sound. There are functions in SDL_sound 2
      for allocating these from a pool, which reduces allocation overhead
      and memory fragmentation, and eliminates multiple trips to the disk
      when you "read" the sound.

    - Sound_Sample structures are allocated in a pool, too, so throwaway
      sounds (specifically, ones using pooled RWOPS) don't thrash system
      resources. Good for those fire-and-forget effects.

    - It's trivial to roll a reference-counting RWOPS that lets you use the
      same memory block for several playing sounds, and when the last one
      closes it (all related Sound_Samples go through Sound_FreeSample()), it
      deletes the original memory block. Handy if you only want to loosely
      manage those buffers.

    - Cull samples if you're playing too many. The app can decide which sounds
      are important and assign them a priority, and let only the first X
      highest priority sounds actually mix.

    - Alternately, if you can swallow the memory: take a highly-compressed
      file and put it into a Sound_Sample, call Sound_DecodeAll. Now, use
      the sample's "decoded" field as raw PCM for other Sound_Samples using
      above tricks. When you are done, clean up the other samples first, then
      call Sound_FreeSample() on this one. This is extremely useful if you
      want to reduce CPU usage for one sound that is otherwise compressed.
      Memory usage doesn't grow exponentially with each simulataneous mixing
      of this sound, because everyone is feeding from the same memory block,
      so each new sample instance adds some bytes for the structures (which
      might have been allocated in the pool already anyhow).


Hardcore #2: Predecoding and buffer sizes.

Take advantage of the 1.0 API for predecoding and altering the decode buffer
size. This gives you control over the memory/CPU tradeoff at mix time, as the
mixer will call Sound_Decode() when it needs more data from a playing sample.
How much decoding is done at that point depends on how much buffering is
available. If you predecode the whole thing with Sound_DecodeAll(), then the
mixer can focus on mixing and not spend time decoding.



Hardcore #3: Global gain, global fade.

Most attributes that apply to one sample can be applied to all by passing a
NULL for the first argument to Sound_SetAttribute(). Gain and fade are
examples of this. If you want everything to fade out at once, this is the
best, race-condition free way to do it.

Note that global attributes don't override (or overwrite) per-sample
attributes. If you set a sample's gain to 2.0 and the global gain to 0.5, the
sound plays at normal (1.0) gain...the sample's gain is still 2.0 when you
change the global gain thereafter.


Hardcore #4: Postmix callbacks.

You can register a callback function per-sample that is called when the mixer
has finished its work and is about to send the data to the audio hardware.
These generally run in seperate threads on most platforms, and as such must be
protected from your main code with mutexes.

These are useful if you are are writing a media player and want to show some
sort of visual feedback based on the playing audio data, or if you want to
tweak the sound with your own special effects.


Hardcore #5: Sample finished callback.

You can register a callback function per-sample that is called when the mixer
has completely finished mixing a non-looping sample. This is largely a nod to
SDL_mixer, where this was the most convenient way to clean up fire-and-forget
sounds, but most people will want to let SDL_sound handle those. This has
other good uses: it lets you know when sound events are complete if you are
adding cinematics/cut-scenes to your program, or perhaps when it's safe for
characters to speak again (it's strange when one actor is speaking two
overlapping lines of dialogue, for example).


Hardcore #6: Procedural samples.
!!! WRITEME: Hook up a RAW Sound_Sample to a RWOPS that generates sound on
!!! WRITEME:  read, so you can mix in a procedural sample instead of just
!!! WRITEME:  pre/post mixing.



Hardcore #7: Your suggestion here!

The goal is to try and make audio work fairly painless for the game developer,
which means that if there is a good way to generalize functionality into the
mixer layer, we probably should. Comments are welcome!

It's worth noting that this tutorial covers common usage patterns and the Big
Important Things, so a lot of support API isn't covered here. For example,
important things like being able to query sample attributes weren't important
enough to mention here, but that doesn't mean you can't do it).

// end of mixer.txt ...