Hacker Guide/Audio Output

Description

An audio output module is used to pass decoded audio frames to the audio output hardware. Normally, the audio frames contain an integral number of audio samples coded linearly, i.e. PCM. However, to address the need for digital audio pass-through, the audio output may also receive coded audio frames encapsulated as S/PDIF packets.

This describes the fourth version of the audio output layer, found in VLC version 1.2.0. Versions 0.5.0 to 1.1.x were based on a partly similar but nevertheless incompatible interface known as aout3.

Writing an audio output module

For more details on writing modules in general, see Documentation:Hacker's Guide/Module Writers Guide. An audio output module should be declared with the correct capability and category as follows:

set_capability("audio output", 60)
set_category(CAT_AUDIO)
set_subcategory(SUBCAT_AUDIO_AOUT)
set_callbacks(Open, Close)

(Note: The callbacks name can be different. We use Open() and Close() in this example.)

Instances of an audio output module should store their private internal state (if any) as an aout_sys_t structure:

struct aout_sys_t
{
    /* ... */
};

Priority

Unless a specific module has been provisioned in the VLC preferences, or specified through the --aout command line parameter, LibVLC will try to each audio output module in order of decreasing priority until one succeeds. As usual modules with a priority of zero will never be probed unless configured explicitly.

Thus, on Linux for instance, the PulseAudio module has a higher priority than the ALSA one, and the OSS module has the smallest (non-zero) priority. The file audio output has zero priority and will not be used normally. This is so that VLC uses PulseAudio if available, otherwise ALSA, otherwise OSS.

Initialization

The audio output will be probed and initialized with the typical Open() callback. The audio output should check the input audio format in audio_output_t.format:

static int Open (vlc_object_t *obj)
{
    audio_output_t *aout = (audio_output_t *)obj;
    aout_sys_t *sys;
    
    /* Check input format */
    vlc_fourcc_t format = aout->format.i_format;
    unsigned samplerate = aout->format.i_rate;
    unsigned channels = aout_FormatNbChannels(&aout->format);
    
    /* Initialize audio subsystem */ 
    sys = malloc (sizeof (*sys));
    if (unlikely(sys == NULL))
        return VLC_ENOMEM;
    /* ... */
    if (failure)
    {
        free (sys);
        return VLC_EGENERIC;
    }
    
    /* Adjust format to stereo */
    aout->format.i_original_channels =
    aout->format.i_physical_channels = AOUT_CHAN_LEFT|AOUT_CHAN_RIGHT;
    /* aout->format.i_format = VLC_CODEC_S16N; */
    /* aout->format.i_rate = 48000; */
    
    /* Setup callbacks */
    aout->sys = sys;
    aout->pf_play = Play;
    aout->pf_pause = Pause;
    aout->pf_flush = Flush;
    aout_VolumeSoftInit (aout);
    return VLC_SUCCESS;
}

When done, the audio output module will be deinitialized:

static void Close (vlc_object_t *obj)
{
    audio_output_t *aout = (audio_output_t)obj;
    aout_sys_t *sys = aout->sys;
    
    /* Deinitialize */
    /* ... */
    
    free (sys);
}

Audio format

The initial format when Open() starts, is usually VLC_CODEC_FL32 (float) on desktop systems, but not always. All formats are implicitly channel-interleaved. Uninterleaved audio is never used. For instance, stereo audio is ordered as:

• first left channel value,
• first right channel value,
• second left channel value,
• second right channel value,
• etc.

Currently, the following formats are supported:

FOURCC	Description	C type	Endianess	On entry	On return
VLC_CODEC_FL32	Single precision	float	Native	With FPU	Allowed
VLC_CODEC_FI32	Fixed-point	int32_t	Native	Without FPU	Forbidden
VLC_CODEC_S16N	Signed 16-bits	int16_t	Native	Without FPU	Allowed
VLC_CODEC_F32L	Single precision	float	Little	With little endian FPU	Allowed
VLC_CODEC_F32B	Single precision	float	Big	With big endian FPU	Allowed
VLC_CODEC_F64L	Double precision	double	Little	Never	Allowed
VLC_CODEC_F64B	Double precision	double	Big	Never	Allowed
VLC_CODEC_S16L	Signed 16-bits	int16_t	Little	Without FPU	Allowed
VLC_CODEC_S16B	Signed 16-bits	int16_t	Big	Without FPU	Allowed
VLC_CODEC_S24L	Signed 24-bits	N/A	Little	Never	Allowed
VLC_CODEC_S24B	Signed 24-bits	N/A	Big	Never	Allowed
VLC_CODEC_S32L	Signed 32-bits	int32_t	Little	Never	Allowed
VLC_CODEC_S32B	Signed 32-bits	int32_t	Big	Never	Allowed
VLC_CODEC_A52	AC-3 / Dolby	Non-linear	N/A	A52 input	Forbidden
VLC_CODEC_DTS	DTS Coherent Acoustics	Non-linear	N/A	DTS input	Forbidden
VLC_CODEC_MPGA	MPEG 2 Audio	Non-linear	N/A	MPEG input	Forbidden
VLC_CODEC_SPDIFL	S/PDIF	uint16_t	Little	Never	Allowed (S/PDIF)
VLC_CODEC_SPDIFB	S/PDIF	uint16_t	Big	Never	Allowed (S/PDIF)

The "on entry" column indicates in which situation (if any), the format may be presented by the LibVLC core to the audio output Open() callback. The "on return" column determines if the format is supported for actual output.

If the returned format differs from the one specified by LibVLC core upon entry, then LibVLC will insert any required conversion filter automatically. So do not convert manually in the output module!

S/PDIF

If and only if the format on entry is non-linear, the audio output module can enable digital pass-through mode. To do so, it must set the audio output format to VLC_CODEC_SPDIFL (or VLC_CODEC_SPDIFB). If on the contrary pass-through is not to be used, then the format must be set to a linear FOURCC, usually VLC_CODEC_FL32 or VLC_CODEC_S16N.

Convenience aliases

VLC_CODEC_FL32 is an alias for VLC_CODEC_F32L or VLC_CODEC_F32B depending on architecture endianess. The full list of convenience native endianess FOURCC is

• VLC_CODEC_FL32 (float)
• VLC_CODEC_FL64 (double)
• VLC_CODEC_S32N (int32_t)
• VLC_CODEC_S24N (N/A)
• VLC_CODEC_S16N (int16_t)
• VLC_CODEC_U16N (uint16_t)

Sample rate and channels

The sample rate is the sample rate that comes from the decoder or the audio filters, so is the channel mapping.

If possible, it is recommended that the audio output uses a format as close to the input as possible. This is to conversion and loss of quality. Nevertheless, it is often necessary to use different a output format due to hardware limitation: - resample if the aout->format.i_rate is modified, - remix the channels if aout->format.i_physical_channels and/or aout->format.i_original_channels are modified, - convert the sample format is aout->format.i_format is modified.

Warning

Beware that the Open() callback must not change aout->format until it is absolutely certain to return VLC_SUCCESS (0). If it were to change the format and then return an initialization failure, subsequent modules would get corrupt informations about the input format!