Imagine clown here

The SL Sound Library.

By Steve Baker


This library allows one to play, mix and modify sound samples in realtime - with a special view towards sound for interactive systems such as games and simulation.

At present, SL can drive PC-style sound cards from within Windows NT, 95 and 98, Linux and several other UNIX systems. When not running under Windows, SL relies on the Open Sound System (OSS) drivers (formerly known as `VoxWare') - which are generally a standard part of the Linux Kernel and are available on most other UNIX implementations. Notably, the SGI IRIX operating system does not support OSS - so IRIX is supported via the proprietary SGI Audio Library. A future port to the Mac operating system is anticipated.

For PC-based Linux systems using relatively 'standard' SoundBlaster compatible sound cards, there is a companion library 'SM' that allows one to drive the Audio mixer on your sound card. Since the SM library is not easily portable to other systems, it is not considered a part of SL. SM may also be ported to Windows - but it's unlikely that it will be more generally available so it's use is deprecated. SM does not exist for IRIX.


Steve Baker <> wrote the first cut of the library for Intel-based Linux machines and maintains the system and its documentation.

Tom Knienieder ported SL onto the Windows family of operating systems, onto OpenBSD and also onto IRIX as a part of the FlightGear project.

Curt Olson is the leader of the FlightGear project and provided useful testing and helpful comments in appropriate quantities.


The SL library allows one to operate at several levels.

Using the Library.

To use this library, you must '#include "sl.h"' and link to libsl.a or

class slDSP.

Hardly any programs will ever create more than one slDSP object - in fact, most programs won't explicitly create any slDSP's at all since they'll be more likely to use slScheduler - which is derived from slDSP.

Here are the member functions:

  class slDSP

    slDSP ( int rate, int stereo, int bps ) ;
    slDSP ( char *device, int rate, int stereo, int bps ) ;
   ~slDSP () ;
    void play ( void *buffer, size_t length ) ;

    int working     () ;  /* For optimists  */
    int not_working () ;  /* For pessimists */

    int getBps   () ;
    int getRate  () ;
    int getStereo() ;

    void sync () ;
    void stop () ;
  } ;

By default, "/dev/dsp" (or just plain "dsp" under Windows) is the device used by this class, but an alternative device may be accessed if it's name is passed into the constructor. The constructor function also needs to know the desired sample replay rate, whether the sounds should be played in stereo (TRUE) or mono (FALSE) and how many bits per pixel of audio are to be provided (typically 8 or 16).

The constructor function cannot guarantee to provide the exact settings you require since hardware sample replay units vary considerably in capabilities from one manufacturer to another. Hence, all applications should call the slDSP::getBps/getStereo/getRate() commands after the object has been created. These routines tell the application the actual number of bits per sample, the stereo setup and the sample rate that this channel is currently set up to support. The getRate command gets the sampling rate in Hz.

The 'stop' function aborts any sounds that are currently being played and the 'sync' function blocks until the current audio stream is flushed.

The 'slDSP::not_working()' function returns TRUE if there is any kind of problem with the driver. Mostly it returns TRUE if your system isn't set up for audio. 'slDSP::working()' is provided by popular request and is just the opposite.

The driver will not fail if you call it's member functions when 'slDSP::not_working()' returns TRUE - so programs written for audio should work OK even on hardware with no audio support.

The play command takes raw audio data from a buffer in memory (typically from the buffer of an slSample) and plays it with the current settings. In 8bps mono mode, the data is a simple unsigned byte stream. In 16bps mode, the data is a stream of unsigned shorts and in stereo mode, samples alternate between left and right channels - left first, then right.

Hence, each sound sample can be 1, 2 or 4 bytes long.

 8bps Mono  :      L      |      L      |...
16bps Mono  :    H   L    |    H   L    |...
 8bps Stereo:    LL LR    |    LL LR    |...
16bps Stereo: HL LL HR LR | HL LL HR LR |...

Where: L is a low  order byte (monophonic),
       H is a high order byte (monophonic),
      LL is low-order, left channel,
      LR is low-order, right channel,
      HL is high-order, left channel,
      HR is high-order, right channel.

class slSample

This class is intended to hold all the relevent information that goes along with a digital sound sample. Most programs will create a separate slSample for each sound effect or musical note:

class slSample

  slSample () ;
  slSample ( Uchar *buff, int leng ) ;
  slSample ( char *fname ) ;
  slSample ( char *fname, slDSP *match ) ;
 ~slSample () ;
  char *getComment () ;
  void   setComment ( char *nc ) ;

  Uchar *getBuffer () ;
  int    getLength () ;
  void   setBuffer ( Uchar *buff, int leng ) ;

  void setRate   ( int r ) ;
  void setBps    ( int b ) ;
  void setStereo ( int s ) ;

  int  getRate   ()        ;
  int  getBps    ()        ;
  int  getStereo ()        ;

  int  getPlayCount ()     ;

  float getDuration ()     ;

  int loadFile    ( char *fname ) ;
  int loadRawFile ( char *fname ) ;
  int loadAUFile  ( char *fname ) ;
  int loadWavFile ( char *fname ) ;

  void changeRate   ( int r ) ;
  void changeBps    ( int b ) ;
  void changeStereo ( int s ) ;

  void autoMatch ( slDSP *player ) ;
  void adjustVolume ( float vol ) ;

  void print ( FILE *fd ) ;
} ;

There are three ways to construct a slSample - first, you can create an (initially) empty sample:

  slSample::slSample () ;

Secondly, you can provide a buffer of audio data (with either 1, 2 or 4 bytes per sample) - along with the length of that buffer (in bytes - NOT in samples).

  slSample::slSample ( Uchar *buff, int leng ) ;

You can also load a file from disk:

  slSample::slSample ( char *fname ) ;

In that case, slSample will look at the extension of the filename and choose an appropriate loader for that file type.

Since sample files from disk may well not match the settings on the slDSP that will be replaying it, there is an option to pass the address of the slDSP that will be doing the replaying and have the slSample automatically match it's settings to that of the replay unit.

You can load a sample file into a slSample at any time using:

  int slSample::loadFile    ( char *fname ) ;
  int slSample::loadRawFile ( char *fname ) ;
  int slSample::loadAUFile  ( char *fname ) ;
  int slSample::loadWavFile ( char *fname ) ;


  void slSample::autoMatch ( slDSP *player ) ;

The basic 'slSample::loadFile' call will auto-detect the file type - the remaining three calls presume the file to be of the type indicated irrespective of the extension.

  Raw - A simple file containing just the samples in the same
        format as required by the slDSP class (see above). These
        often have the file extension '.ub' (unsigned byte).
        Since this format has no header, it is always assumed to be
        mono, 8000Hz and 8bps. If you know better - then use
        setRate/setStereo/setBsp to sort out the resulting mess!
  AU  - (Sun Microsystems 'audio' format), this file type typically
        has the '.au' extension.
  Wav - (Microsoft 'WAVE' format), this file type typically
        has the '.wav' extension.

All flavours of 'slSample::loadFile' return FALSE if the file cannot be read, TRUE otherwise. slSample::autoMatch allows the sample to be tweaked to work correctly with the slDSP that will be used to replay the sample subsequently.

  char *slSample::getComment () ;
  void  slSample::setComment ( char *nc ) ;

Each sound sample can have a comment string associated with it. Both '.wav' and '.au' files support such strings. In a 'wav' file, there may be multiple strings - only the first is read.

  Uchar *slSample::getBuffer () ;
  int    slSample::getLength () ;
  void   slSample::setBuffer ( Uchar *buff, int leng ) ;

These calls allow low level access to the sample buffer contained in an slSample. slSample::setBuffer makes a private copy of the buffer you pass to it, so you may delete your copy afterwards. slSample::getBuffer returns a pointer to the local buffer inside the slSample - you must not free it up. If you want to free up the memory in a slSample without deleting it then you should call slSample::setBuffer(NULL,0).

The following calls get and set the internal parameters of an slSample:

  void slSample::setRate   ( int r ) ;
  void slSample::setBps    ( int b ) ;
  void slSample::setStereo ( int s ) ;

  int  slSample::getRate   ()        ;
  int  slSample::getBps    ()        ;
  int  slSample::getStereo ()        ;

Note that setting a parameter does not affect the audio data in the buffer. If a sample was recorded at (say) 8KHz and you call slSample::setRate(16000), then the sample will replay at 16KHz but will be only half the usual duration and twice the frequency. The effects of slSample::setBps and slSample::setStereo are even more destructive if used inappropriately.

The main use for the slSample::setRate/setBps/setStereo are for when the audio data was created inside the program or read from a 'Raw' file format and there was no other indication of these settings. By default, such samples are assumed to be 8KHz, monophonic 8bps recordings. This is supported by almost all audio cards.

If you need to manipulate the sample so that it sounds right when replayed with settings that differ from those when it was recorded then you need:

  void slSample::changeRate   ( int r ) ;
  void slSample::changeBps    ( int b ) ;
  void slSample::changeStereo ( int s ) ;

It is sometimes useful to permenantly change the volume of a sound sample:

  void slSample::adjustVolume ( float vol ) ;

Setting vol to 2.0 would double the volume, setting it to 0.5 would halve it.

Bear in mind though that these change/adjust routines make permenant changes to the data in the sound buffer. The more changes you make, the more noise you'll introduce into the sample. If possible, record 16 bit samples and play with them in a proper sample studio package - then reduce them to 8 bits before you load them into an slSample.

These routines are also relatively slow - they create a new sound buffer, process the old data to fit the new setup and then free up the old buffer. The length and address of the internal slSound buffer will often change as a result.

  float slSample::getDuration () ;

This returns the duration of the sound sample in seconds.

  int slSample::getPlayCount () ;

Returns a count of the number of simultaneous instances of this sample are currently being played by slSchedulers (see below).

  void slSample::print ( FILE *fd ) ;

This prints out all the parameters of the sound sample to the specified file descriptor in human-readable form.

class slScheduler.

This class is where it all comes together. Most programs will only create a single slScheduler. Since the slScheduler needs dedicated access to the DSP, this class is inherited from an slDSP - so most programs that use slScheduler should not declare an slDSP as well.

IMPORTANT NOTE: Since slScheduler is derived from slDSP, all slDSP member functions are available as slScheduler calls.

class slScheduler : public slDSP
  slScheduler ( int rate ) ;
  slScheduler ( char *device, int rate ) ;
 ~slScheduler () ;

  float setSafetyMargin ( float seconds ) ;

  void update () ;
  void dumpUpdate () ;

  void stopSample   ( slSample *s, int magic ) ;
  void pauseSample  ( slSample *s, int magic ) ;
  void resumeSample ( slSample *s, int magic ) ;

  int loopSample ( slSample *s, int pri = 0,
                   slPreemptMode mode = SL_SAMPLE_MUTE,
                   int handle = 0, slCallBack cb = NULL ) ;
  int playSample ( slSample *s, int pri = 1,
                   slPreemptMode mode = SL_SAMPLE_ABORT,
                   int handle = 0, slCallBack cb = NULL ) ;

  void stopMusic   ( int magic ) ;
  void pauseMusic  ( int magic ) ;
  void resumeMusic ( int magic ) ;

  int loopMusic  ( char *fname, int pri = 0,
                   slPreemptMode mode = SL_SAMPLE_MUTE,
                   int handle = 0, slCallBack cb = NULL ) ;
  int playMusic  ( char *fname, int pri = 1,
                   slPreemptMode mode = SL_SAMPLE_ABORT,
                   int handle = 0, slCallBack cb = NULL ) ;
} ;

The idea behind slScheduler is that it manages all your music and sound sample replay needs. You construct it at the start of your program run, load in your slSamples and then call the 'slScheduler::update' function every iteration of your code.

The constructor function accepts a sample rate argument - note the caveats in the slDSP contructor function - you need to call getRate() to determine if your hardware was actually able to support the rate you asked for - or whether you actually obtained some slightly different rate. If you don't get a rate that matches your sound samples then you'll need to call slSample::changeRate() or even slSample::autoMatch for each one in turn to change it's sampling rate to something that suits the hardware.

Notice that at present, slScheduler only supports 8bps monophonic playback, this may change in the future.

Once everything is up and running, you can very easily play either slSamples or music. Simply call:

  int slScheduler::loopSample ( slSample *s ) ;
  int slScheduler::playSample ( slSample *s ) ;

  int slScheduler::loopMusic ( char *fname ) ;
  int slScheduler::playMusic ( char *fname ) ;

There are additional arguments to these calls - but they are all optional. slScheduler::loopSample() tells the scheduler to play that sample in a loop forever, the slScheduler::playSample() tells it to play the sample as a single one-shot effect.

Similarly, you can read and play music (currently only in 'MOD' format). slScheduler::loopMusic() tells the scheduler to play the music in a loop forever, the slScheduler::playMusic() tells it to play the file as a single one-shot playback.

You can only play back one music track at a time.


The slScheduler can (at present) play only three sounds at the same time. You can ask it to play more than that - but only three will actually sound at any given instant. If too many sounds try to play at once, all but three of them will be 'pre-empted'.

You can tag each sample with a priority and a 'pre-empt' mode. The priority indicates which sounds have priority over which others and the 'pre-empt' mode tells it what to do if a sound of higher priority wants to play in the meantime.

The full form of the slScheduler::loopSample/Music and slScheduler::playSample/Music calls is:

  int slScheduler::loopSample ( slSample *s, int priority, slPreemptMode mode ) ;
  int slScheduler::playSample ( slSample *s, int priority, slPreemptMode mode ) ;
  int slScheduler::loopMusic  ( char *fname, int priority, slPreemptMode mode ) ;
  int slScheduler::playMusic  ( char *fname, int priority, slPreemptMode mode ) ;

The 'priority' is a simple integer. Zero is the least important sound, SL_MAX_PRIORITY (currently set at 16) is the most important.

The 'mode' parameter lets you choose what happens to a sound when it is pre-empted. 'mode' is one of:

Interacting with a Sample that's Playing.

Also, you can ask the scheduler to tell you when something important happens to your sound.

  int slScheduler::loopSample ( slSample *s, int priority, slPreemptMode mode,
                                int magic, slCallBack cb ) ;
  int slScheduler::playSample ( slSample *s, int priority, slPreemptMode mode,
                                int magic, slCallBack cb ) ;
                                int magic, slCallBack cb ) ;
  int slScheduler::loopMusic  ( char *fname, int priority, slPreemptMode mode,
                                int magic, slCallBack cb ) ;
  int slScheduler::playMusic  ( char *fname, int priority, slPreemptMode mode,
                                int magic, slCallBack cb ) ;

The 'cb' parameter is the address of a callback function that SL will call whenever something important happens to your sound.

Your function much accept three parameters:

  typedef void (*slCallBack) ( class slSample *sample, slEvent event, int magic ) ;

The first parameter is the slSample that was being played when this even happened, the second parameter tells you what happened:

  SL_EVENT_COMPLETE  -- The sound finished playing.
  SL_EVENT_LOOPED    -- The sound looped back to the start.
  SL_EVENT_PREEMPTED -- The sound was preempted by another sound.

Other events may be added in the future, so your callback function should be prepared to do nothing gracefully if an unknown event comes along.

Since there may be multiple copies of a single sample being played at the same time, the 'magic' parameter is a means for you to identify the context that the sound comes from. You pick a magic number that's meaningful to your application when you call slScheduler::playSample or slScheduler::loopSample - and that number will be passed back to your callback function whenever it is called. SL doesn't do anything else with this magic number - so it's meaning can be anything you want provided that the magic number must be non-zero.

A callback function can start new sounds playing, but if you wait for the SL_EVENT_COMPLETE event to do that then there may be a fraction of a second pause between the sound that just stopped and the new one starting.

There are other ways to use the magic number to interact with playing samples:

   void slScheduler::stopSample   ( slSample *sample, int magic ) ;
   void slScheduler::pauseSample  ( slSample *sample, int magic ) ;
   void slScheduler::resumeSample ( slSample *sample, int magic ) ;

   void slScheduler::stopMusic   ( int magic ) ;
   void slScheduler::pauseMusic  ( int magic ) ;
   void slScheduler::resumeMusic ( int magic ) ;

This stops/pauses/resumes all instances of the specified sample with the specified magic number. The 'stopSample' command will even stop looped samples. If the 'sample' is NULL then all samples with that magic number are stopped/paused/resumed. If the 'magic' number is zero then all instances of 'sample' are affected. If sample is NULL and magic is zero then all currently playing sounds are affected.

IMPORTANT NOTE: When you are playing a sample, it is important not to delete it until after it has finished playing. SL keeps count of the number of times each sample is playing - and will produce a fatal error and exit if you attempt to do this. Note that even if you tell the sample to stop playing using slScheduler::stopSample, you must not delete it until *AFTER* the next call to slScheduler::update(). You can use the slSample::getPlayCount() function to find out how many playing instances of a particular slSample there are.


Most of the time, a sound can be simply played as-is. This is highly desirable since doing real-time audio modification is costly. However, there are times when you really need to alter the pitch, volume, stereo-postion (pan) or filtering of a sample in realtime.

One classic example of this is in a car racing game when you want the pitch and volume of the engine to change in response to the throttle and gear ratio - even though the sound itself is really a simple looped sample.

To cater for these needs, SL has the concept of an 'envelope'. An envelope is a array of floating point values and and array of times at which those values are correct. The envelope can be attached to a playing sound like this:

   slScheduler::addSampleEnvelope ( slSample *s, int magic,
              int slot, slEnvelope *e, slEnvelopeType type ) ;
   slScheduler::addMusicEnvelope ( int magic,
              int slot, slEnvelope *e, slEnvelopeType type ) ;

As with the other sample interaction commands, the sample and magic number can be used as wild-cards to apply the same command to a number of sample instances.

Each sample instance can has a minimum of four (and a maximum of SL_MAX_ENVELOPES) slots into which an envelope can be placed. Each envelope can operate on a different aspect of the sample and the envelopes are applied to the sound in slotwise order.

To remove an envelope from a particular sample/magic/slot combination, simply use slScheduler::addSampleEnvelope or slScheduler::addMusicEnvelope with a NULL envelope.

Using a lot of envelopes will considerably increase the amount of CPU time consumed by SL - so beware of trying overly fancy effects.

The 'slEnvelopeType' argument defines which aspect of the sample's performance will be affected by the envelope:

  SL_PITCH_ENVELOPE  : The value in the envelope determines the
                       rate at which the sample is speeded up
                       or slowed down during replay. A value of
                       2.0 would double the pitch of the sound and
                       halve it's duration. A value of 0.5 would
                       halve the pitch and double the duration.

  SL_VOLUME_ENVELOPE : The value of the envelope multiplies the
                       amplitude (volume) of the sound. Beware of
                       large multipliers that would cause the sound
                       to be clipped and small multipliers that
                       increase the noise level of the sample.

                       The value of this parameter determines the
                       number of consecutive samples that are
                       averaged together in a kind of moving average.
                       Large numbers tend to muffle the sound.

                       The stereo position of the sound is moved
                       to the left (negative numbers) or to the
                       right (positive numbers). +/-1.0 is full-scale.

                       The sound is added to a half volume copy of
                       itself which is shifted in time by this number
                       of seconds.

In addition to these, there are a set of SL_INVERSE_PITCH_ENVELOPE, SL_INVERSE_VOLUME_ENVELOPE, ...etc. These operate in an identical manner to the standard envelopes - but with the opposite sense. Hence, if you want to cross-fade two similar sounds then apply the same envelope as an SL_VOLUME_ENVELOPE to one sample and as an SL_INVERSE_VOLUME_ENVELOPE to the other and they will neatly cross-fade.

Here is the class slEnvelope:

class slEnvelope
  slEnvelope ( int _nsteps = 1, slReplayMode _rm,
                                float *_times, float *_values ) ;
  slEnvelope ( int _nsteps = 1, slReplayMode _rm ) ;
 ~slEnvelope () ;

  int getPlayCount () ;

  void setStep ( int n, float _time, float _value ) ;

  float getStepValue ( int n ) ;
  float getStepTime  ( int n ) ;

  float getValue ( float _time ) ;
} ;
The envelope can be constructed either with or without initial data - and can be either SL_SAMPLE_ONE_SHOT or SL_SAMPLE_LOOP. A one-shot envelope retains it's final value indefinitely after it finishes playing - a looped envelope returns to the start and repeats indefinitely.

One very common thing to do is to create an slEnvelope containing just one value.

The slEnvelope::setStep() function allows you to modify the value of the n'th step of the envelope and to position it in time. You can use this in realtime to dynamically interact with the envelope and hence modify whatever samples are using that envelope at the time. You can read back the time and value for any given step of the envelope using slEnvelope::getStepTime() and slEnvelope::getStepValue().

Envelopes interpolate between the steps you provide - so if a pitch envelope had just three value/time pairs:

  slEnvelope my_envelope ( 3, SL_SAMPLE_LOOP ) ;
  my_envelope . setStep ( 0,  0.0, 1.0 ) ;
  my_envelope . setStep ( 1, 10.0, 2.0 ) ;
  my_envelope . setStep ( 2, 20.0, 1.0 ) ;

  scheduler -> playSample ( my_sample ) ;
  scheduler -> addSampleEnvelope ( my_sample, 0,
                     0, & my_envelope, SL_PITCH_ENVELOPE ) ;

...then this could be applied to a sample which would then gradually increase in pitch over the next 10 seconds - by which point, the pitch would have doubled - and over the following 10 seconds, would gradually return to normal. This behaviour would repeat for as long as the envelope remained attached to that sample.

slEnvelope::getValue() returns the interpolated value at the specified time.

You can remove an envelope from a sample thats playing by passing a NULL envelope in the same slot.

Timing the Audio Updates.

The slScheduler has to pump audio data into the Linux device driver in realtime. It does this by transferring a chunk of pre-mixed audio data over to the Linux device driver every time the it seems to be getting low on data.

Note that libsl is not using a separate thread to do this - although you could place calls to slScheduler::update() in a separate thread if you wanted to. That means that if you call the update function too infrequently, the Linux device driver will run out of data and you will hear breaks and clicks in the audio.

Here is some advice on how to deal with this problem. slScheduler supports this call:

  float slScheduler::setSafetyMargin ( float num_seconds ) ;

The slScheduler::setSafetyMargin() call allows you to tell the scheduler to ensure that there are at least 'num_seconds' of audio queued up in the driver after each call to update. (The default safety margin is two seconds - which is likely to be on the long side in many applications).

If you make this number very large (say 4 seconds or more) then Linux won't have enough buffer space to hold that amount of audio data in it's internal buffer (which is about 64Kb long on my system) - and slScheduler::update() will block indefinitely trying to keep it that full. The audio will be smooth - but your application will get zero time to run in!

If you set the safety margin to a smaller number (but still much larger than the rate that you call update - say 1 or 2 seconds) then the system will produce smooth, continuous audio - but there will be a significant delay between calling the slScheduler::playSample() function and that sample actually playing. That is because there is still a number of seconds of data in the Linux buffer that has to be played before your new sound will be heard.

For music replay, this is fine - all the audio was pre-planned, so it'll sound great. But if you are creating the audio for a game or something - then it's probably important that the "BLAM!!" sound happens very soon after the player presses the fire button. With safetyMargin set to two seconds, you could get anywhere up to a two second delay!

Here is one (partial) remedy:

If you know that a VERY important sound has to be played RIGHT NOW - then you can call the slScheduler::dumpUpdate() function instead of the usual slScheduler::update() on that frame. This will have the effect of telling Linux to dump all the audio that has been queued but not yet played - so that your important sound can play RIGHT NOW. The downside is that a significant amount of sounds that were sitting in the Linux queue will be dumped - and a certain amount of CPU time will be wasted as a result of having to compute those seconds of audio twice.

If you were playing a boring buzzing noise for the sound of an engine in a flight simulator - then losing a second or two of it might be acceptable. If you were playing continuous music in the background then using slScheduler::dumpUpdate() would sound like your CD player skipping when you jog it too hard.

The final solution is to set safetyMargin to something quite small (say 0.2 seconds). This means that in theory, all your sounds will be heard within 0.2 seconds of you asking for them to play. Of course if you should fail to call the slScheduler::update() function at LEAST once every 1/5th second, Linux could run out of audio data and you'd get gaps and clicks which sound REALLY TERRIBLE.

I find that even with the most simple application, I get bad breakup with a safetyMargin anywhere under 0.13 seconds - but that's on a fairly ancient 100MHz 486 - I would expect a modern CPU to do better.

The sound library can keep up with an 8KHz mono, 8bps sample rate consuming only 1ms of CPU time every iteration - also on my 100MHz 486.

There are times when you might want to change safetyMargin on-the-fly. This is fine - but the effect of the change won't be noticable until the data that is currently queued within Linux has been played.

The Volume Level Conundrum.

Globally, the volume is controlled by the audio mixer. If you are using the 'SM' companion library then you can do this with class slMixer - no problem there - but the relative volume of the samples that are played together cannot be controlled so easily.

For speed, the scheduler simply adds the sounds together and clamps any values that would overflow. To do anything more complex would consume significant amounts of CPU time.

(Since sounds are stored as unsigned char's the value 0x80 actually represents zero voltage - so the equation for adding two sound samples 'a' and 'b' is:

   a + b - 0x80 

it is important to bear this in mind if you do any work on the slSample buffers yourself.)

Clearly if two sounds each range in voltage from utter minimum (0x00) to absolute maximum (0xFF) then there will be considerable distortion when they are added. The problem gets even worse when there are three sounds. Hence, if you really expect to play two sounds at once then you should reduce the volume of each sample to the range 0x40 to 0xC0, and if you expect to play three sounds at once then the range 0x56 to 0xAA. However, if you do that then you are in effect losing one or two bits of audio precision - which raises noise levels quite a bit.

It would have been possible to go to 16 bit sound samples - but not all PC sound cards can do that - and in any case, it would double the amount of memory and CPU time taken to play the sounds.

One notable fact is that it is fairly unlikely that all your sounds actually fill the range 0x00 to 0xFF exactly. Most sampled sounds that you find recorded by home equipment are typically either:

If under-recorded, then further reducing the volume for the sake of adding several sounds together may be unnecessary.

If over-recorded, then you'll need to reduce the volume of the sample for sure since it's already being clipped.

Even in the happy case of perfect recordings, it's pretty unlikely that the loudest parts of two sounds will happen to replay at the exact same instant - and even if they did, it might be in some rare spikes of audio where quality might not suffer.

So, there is no solid mechanism to ensure the best quality. You might want to spend some time playing with the relative volumes of the various samples you use in your code. The slSample::changeVolume() function is too slow to use in realtime - but it is handy for tuning the samples right after they have been loaded from disk.

The Future.

I keep hearing about compressed '.WAV' files - I havn't found any yet - or a spec on how to read them.

There doesn't seem to be a spec for '.AU' files either - but the header format seems simple enough from just looking at a few example files.

It's also possible to allow more simultaneous sounds - the limit of three is a bit arbitary - but again, CPU costs are something to be concerned about. Fortunately, if I ever increase the limit, it shouldn't affect existing programs.

It would be nice if the scheduler knew how to do stereo and 16 bps.

A Note about Reference Counting.

Instances of the slEnvelope and slSample classes are 'reference counted' within SL. That means that a counter in each object is incremented each time you start one playing - and decremented each time one stops. When you 'delete' an slEnvelope or an slSample, the reference count will be checked to ensure it is zero - if it is not then a fatal error will be produced if you then attempt to call slScheduler::update() since that would play a sound that has been deleted by the application.

The error does not occour at the point when you delete the sample or envelope since it is OK to do that so long as you never try to play more sounds later. It is therefore illegal to delete a sample or envelope and then stop it playing.

This arrangement allows an application to exit with outstanding sounds queued up.

Example Program.

This example creates an 'engine' sound in memory using some summed sine waves, loads a number of sounds (and has them automatically matched to the current sound replayer), then plays the engine sound in a loop, periodically interrupting it with some other noises.

#include "sl.h"
#include "sm.h"
#include <math.h>

  Construct a sound scheduler and a mixer.

slScheduler sched ( 8000 ) ;
smMixer mixer ;

int main ()
  mixer . setMasterVolume ( 30 ) ;
  sched . setSafetyMargin ( 0.128 ) ;

  /* Just for fun, let's make a one second synthetic engine sample... */

  Uchar buffer [ 8000 ] ;

  for ( int i = 0 ; i < 8000 ; i++ )
    /* Sum some sin waves and convert to range 0..1 */

    float level = ( sin ( (double) i * 2.0 * M_PI / (8000.0/ 50.0) ) +
                    sin ( (double) i * 2.0 * M_PI / (8000.0/149.0) ) +
                    sin ( (double) i * 2.0 * M_PI / (8000.0/152.0) ) +
                    sin ( (double) i * 2.0 * M_PI / (8000.0/192.0) )
                  ) / 8.0f + 0.5f ;

    /* Convert to unsigned byte */

    buffer [ i ] = (Uchar) ( level * 255.0 ) ;

  /* Set up four samples and a loop */

  slSample  *s = new slSample ( buffer, 8000 ) ;
  slSample *s1 = new slSample ( "scream.ub", & sched ) ;
  slSample *s2 = new slSample ( "zzap.wav" , & sched ) ;
  slSample *s3 = new slSample ( "", & sched ) ;
  slSample *s4 = new slSample ( "wheeee.ub", & sched ) ;

  /* Mess about with some of the samples... */

  s1 -> adjustVolume ( 2.2  ) ;
  s2 -> adjustVolume ( 0.5  ) ;
  s3 -> adjustVolume ( 0.2  ) ;

  /* Play the engine sample continuously. */

  sched . loopSample ( s ) ;

  int tim = 0 ;  /* My periodic event timer. */

  while ( SL_TRUE )
    tim++ ;  /* Time passes */

    if ( tim % 200 == 0 ) sched.playSample ( s1 ) ;
    if ( tim % 180 == 0 ) sched.playSample ( s2 ) ;
    if ( tim % 150 == 0 ) sched.playSample ( s3 ) ;
    if ( tim % 120 == 0 ) sched.playSample ( s4 ) ;

      For the sake of realism, I'll delay for 1/30th second to
      simulate a graphics update process.

#ifdef WIN32
    Sleep ( 1000 / 30 ) ;  /* 30Hz */
    usleep ( 1000000 / 30 ) ;  /* 30Hz */

      This would normally be called just before the graphics buffer swap
      - but it could be anywhere where it's guaranteed to get called
      fairly often.

    sched . update () ;

Valid HTML 4.0!
Steve J. Baker. <>