libvorbisenc release 1.1 - 20040709
Libvorbisenc is an encoding convenience library intended to encapsulate the elaborate setup that libvorbis requires for encoding. Libvorbisenc gives easy access to all high-level adjustments an application may require when encoding and also exposes some low-level tuning parameters to allow applications to make detailed adjustments to the encoding process.
All the libvorbisenc routines are declared in "vorbis/vorbisenc.h". Note: libvorbis and libvorbisenc always encode in a single pass. Thus, all possible encoding setups will work properly with live input and produce streams that decode properly when streamed. See the subsection titled "managed bitrate modes" for details on setting limits on bitrate usage when Vorbis streams are used in a limited-bandwidth environment.
Libvorbisenc is used only during encoder setup; its function is to automate initialization of a multitude of settings in a vorbis_info structure which libvorbis then uses as a reference during the encoding process. Libvorbisenc plays no part in the encoding process after setup.
Encode setup using libvorbisenc consists of three steps:
|sampling rate||The sampling rate (in samples per second) of the input audio. Common examples are 8000 for telephony, 44100 for CD audio and 48000 for DAT. Note that a mono sample (one center value) and a stereo sample (one left value and one right value) both are a single sample.|
|channels||The number of channels encoded in each input sample. By default, stereo input modes (two channels) are 'coupled' by Vorbis 1.1 such that the stereo relationship between the samples is taken into account when encoding. Stereo coupling my be disabled by using vorbis_encode_ctl() with OV_ECTL_COUPLE_SET.|
|quality||A decimal float value requesting a desired quality. Libvorbisenc 1.1 allows quality requests in the range of -0.1 (lowest quality, smallest files) through +1.0 (highest-quality, largest files). Quality -0.1 is intended as an ultra-low setting in which low bitrate is much more important than quality consistency. Quality settings 0.0 and above are intended to produce consistent results at all times.|
Beginning in libvorbis 1.1, bitrate management is implemented using a bit-reservoir algorithm. The encoder has a fixed-size reservoir used as a 'savings account' in encoding. When a frame is smaller than the target rate, the unused bits go into the reservoir so that they may be used by future frames. When a frame is larger than target bitrate, it draws 'banked' bits out of the reservoir. Encoding is managed so that the reservoir never goes negative (when a maximum bitrate is specified) or fills beyond a fixed limit (when a minimum bitrate is specified). An 'average bitrate' request is used as the set-point in a long-range bitrate tracker which adjusts the encoder's aggressiveness up or down depending on whether or not frames are coming in larger or smaller than the requested average point.
|maximum bitrate|| The maximum allowed bitrate, set in bits
per second. If the bitrate would otherwise rise such that oversized
frames would underflow the bit-reservoir by consuming banked bits,
bitrate management will force the encoder to use fewer bits per frame
by encoding with a more aggressive psychoacoustic model.
This setting is a hard limit; the bitstream will never be allowed, under any circumstances, to increase above the specified bitrate over the average period set by the reservoir; it may momentarily rise over if inspected on a granularity much finer than the average period across the reservoir. Normally, the encoder will conserve bits gracefully by using more aggressive psychoacoustics to shrink a frame when forced to. However, if the encoder runs out of means of gracefully shrinking a frame, it will simply take the smallest frame it can otherwise generate and truncate it to the maximum allowed length. Note that this is not an error and although it will obviously adversely affect audio quality, a Vorbis decoder will be able to decode a truncated frame into audio.
|average bitrate||The average desired bitrate of a stream, set in bits per second. Average bitrate is tracked via a reservoir like minimum and maximum bitrate, however the averaging reservior does not impose a hard limit; it is used to nudge the bitrate toward the desired average by slowly adjusting the psychoacoustic aggressiveness. As such, the reservoir size does not affect the average bitrate behavior. Because this setting alone is not used to impose hard bitrate limits, the bitrate of a stream produced using only the average bitrate constraint will track the average over time but not necessarily adhere strictly to that average for any given period. Should a strict localized average be required, average bitrate should be used along with minimum bitrate and maximum bitrate.|
The minimum allowed bitrate, set in bits per second. If
the bitrate would otherwise fall such that undersized frames would
overflow the bit-reservoir with unused bits, bitrate management will
force the encoder to use more bits per frame by encoding with a less
aggressive psychoacoustic model.
This setting is a hard limit; the bitstream will never be allowed, under any circumstances, to drop below the specified bitrate over the average period set by the reservoir; it may momentarily fall under if inspected on a granularity much finer than the average period across the reservoir. Normally, the encoder will fill out undersided frames with additional useful coding information by increasing the perceived quality of the stream. If the encoder runs out of useful ways to consume more bits, it will pad frames out with zeroes.
|reservoir size|| The size of the minimum/maximum bitrate
tracking reservoir, set in bits. The reservoir is used as a 'bit
bank' to average out localized surges and dips in bitrate while
providing predictable, guaranteed buffering behavior for streams to be
used in situations with constrained transport bandwidth. The default
setting is two seconds of average bitrate.
When a single frame is larger than the maximum allowed overall bitrate, the bits are 'borrowed' from the bitrate reservoir; if the reservoir contains insufficient bits to cover the defecit, the encoder must find some way to reduce the frame size.
When a frame is under the minimum limit, the surplus bits are placed into the reservoir, banking them for future use. If the reservoir is already full of banked bits, the encoder is forced to find some way to make the frame larger.
If the frame size is between the minimum and maximum rates (thus implying the minimum and maximum allowed rates are different), the reservoir gravitates toward a fill point configured by the reservoir bias setting described next. If the reservoir is fuller than the fill point (a 'surplus of surplus'), the encoder will consume a number bits from the reservoir equal to the number of the bits by which the frame exceeds minimum size. If the reservoir is emptier than the fillpoint (a 'surplus of defecit'), bits are returned to the reservoir equaling the current frame's number of bits under the maximum frame size. The idea of the fill point is to buffer against both underruns and overruns, by trying to hold the reservoir to a middle course.
Reservoir bias is a setting between 0.0 and 1.0 that biases bitrate
management toward smoothing bitrate spikes (0.0) or bitrate peaks
(1.0); the default setting is 0.1.
Using settings toward 0.0 causes the bitrate manager to hoard bits in the bit reservoir such that there is a large pool of banked surplus to draw upon during short spikes in bitrate. As a result, the encoder will react less aggressively and less drastically to curtail framesize during brief surges in bitrate.
Using settings toward 1.0 causes the bitrate manager to empty the bit reservoir such that there is a large buffer available to store surplus bits during sudden drops in bitrate. As a result, the encoder will react less aggressively and less drastically to support minimum frame sizes during drops in bitrate and will tend not to store any extra bits in the reservoir for future bitrate spikes.
|average track damping||
A decimal value, in seconds, that controls how quickly the average
bitrate tracker is allowed to slew from enforcing minimum frame sizes
to maximum framesizes and vice versa. Default value is 1.5
When the 'average bitrate' setting is in use, the average bitrate tracker uses an unbounded reservoir to track overall bitrate-to-date in the stream. When bitrates are too low, the tracker will try to nudge bitrates up and when the bitrate is too high, nudge it down. The damping value regulates the maximum strength of the nudge; it describes, in seconds, how quickly the tracker may transition from an extreme nudge in one direction to an extreme nudge in the other.
In Vorbis 1.1, vorbis_encode_ctl() can adjust the following additional parameters not described elsewhere:
|management mode||Configures whether or not bitrate management is in use or not. Normally, this value is set implicitly during encoding setup; however, the supported means of selecting a quality mode by bitrate (that is, requesting a true VBR stream, but doing so by asking for an approximate bitrate) is to use vorbis_encode_setup_managed() and then to explicitly turn off bitrate management by calling vorbis_encode_ctl() with OV_ECTL_RATEMANAGE2_SET|
|coupling||Stereo encoding (and in the future, surround encodings) are normally encoded assuming the channels form a stereo image and that lossy-stereo modelling is appropriate; this is called 'coupling'. Stereo coupling may be explicitly enabled or disabled.|
|lowpass||Sets the hard lowpass of a given encoding mode; this may be used to conserve a few bits in high-rate audio that has limited bandwidth, or in testing of the encoder's acoustic model. The encoder is generally already configured with ideal lowpasses (if any at all) for given modes; use of this parameter is strongly discouraged if the point is to try to 'improve' a given encoding mode for general encoding.|
|impulse coding aggressiveness||By default, libvorbis attempts to compromise between preventing wide bitrate swings and high-resolution impulse coding (which is required for the crispest possible attacks, but also requires a relatively large momentary bitrate increase). This parameter allows an application to tune the compromise or eliminate it; A value of 0.0 indicates normal behavior while a value of -15.0 requests maximum possible impulse resolution.|
copyright © 2004 Vorbis team
libvorbisenc release 1.1 - 20040709