Ogg Vorbis: Audio Compression Scheme

1
Ogg Vorbis Audio Compression Scheme

• Presented by Greg Eustace
• MUMT 611

2
Overview

• Introduction
• Licensing
• Availability
• Audio quality metric
• File structure
• Encoding/decoding procedure

3
Introduction

• Vorbis is a lossey audio compression scheme
  intended to replace other similar formats (MP3,
  etc.). It uses a psychoacoustic model to
  eliminate perceptually negligible information,
  thereby decreasing file size.
• The architecture is forward-adaptive, encouraging
  future improvements to the encoding scheme, and
  flexible in that a range of encoding algorithms
  are possible.
• Vorbis website http//www.vorbis.com/

4
Introduction

• Ogg is the transport mechanism in which Vorbis
  audio is contained. It can hold audio, video (Ogg
  Theora) and metadata, alone or multiplexed. Ogg
  is also compatible with other audio formats such
  as FLAC and Speex. Vorbis requires a container
  (such as Ogg) for framing, synchronization,
  positioning and error correction.
• Both Ogg and Vorbis are developed by the Xiph.org
  Foundation
• http//www.xiph.org/

5
Licensing and Availability

• Ogg Vorbis is open source, unpatented and free.
  There are no licensing fees for developers,
  musicians, record labels, etc.
• Encoders/decoders are available and compatible
  with many popular media players on all major
  platforms. Among these are the official command
  line encoder called Oggenc and a graphical
  encoder/decoder/player called OggDrop
• Software can be obtained here http//www.vorbis.c
  om/setup/

6
Audio quality metric

• Ogg Vorbis audio quality is measured on a
  subjective quality scale ranging from -1 to 10.
  The default quality setting is 3, which the
  developers claim sounds better than 128 kbps
  MP3s, while occupying 10 of the file size.
• Although each quality rating corresponds to an
  average bit rate, the encoder does not attempt to
  achieve a specific ABR. Moreover, the developers
  hope to avoid judging quality in terms of this
  criterion. The logic being that bit rates
  correspond only loosely to measures of audio
  quality where differing compression algorithms
  are in use.
• This is demonstrated in comparison between Vorbis
  and MP3, MP3Pro, WMA, ACC and Real Audio, where
  comparable average bit rates clearly correspond
  to varying degrees of audio quality.
• Comparisons can be found here http//www.xiph.org
  /vorbis/listen.html

7
Audio quality metric

• Additional specifications
• Encodes using variable bit rates (VBR) or average
  bit rates (ABR). ABR is used for streaming to
  meet bandwidth requirements.
• Supports up to 255 channels of audio.
• Works with sample rates from 8kHz to 192kHz
• Can potentially support bit rate peeling. This
  entails conversion of an already compressed file
  to a lesser quality without reintroducing (and
  thereby compounding) the same encoding artifacts
  associated with the first conversion.

8
File Structure

• The Vorbis I specification can be found here
• http//www.xiph.org/vorbis/doc/Vorbis_I_spec.html
• The Vorbis bit stream specification consists of
  four packet types, which occur consecutively. The
  first three are headers. The header size is
  unlimited.
• The identification header Identifies the bit
  stream as Vorbis and gives the version in use. It
  includes audio characteristics required for
  further interpretation such as sampling rate and
  channel number.
• The comment header Includes tags consisting of
  user comments and a vendor string. Tags
  themselves may be user-defined.

9
File Structure

• 3. The codec setup header Setup components
  include modes, mappings, floors, residues
  and codebooks, all of which have specific roles
  in the decoding process.
• Codebooks are required for decoding the audio
  stream. For efficiency, audio is represented by
  codewords derived using vector quantisation and
  entropy encoding methods (Huffman binary tree
  representation). Encoding/decoding of individual
  audio packets involves reading from the
  appropriate codebook.
• Audio packets.

10
Encoding Decoding

• In general, encoding involves separating the
  input audio into frames which are compressed to
  form packets. An overlapping transform called the
  Modified Discrete Cosine Transform (MDCT), which
  is a type of discrete Fourier transform, is used
  to convert time domain data to the frequency
  domain for further processing.
• The MDCT involves time-domain aliasing
  cancellation (TDAC) which cancels errors
  resulting from the IMDCT by overlapping (using a
  50 overlap) and adding windows. The decoder
  synthesizes audio frames from the packets and
  reassembles them to approximate the original
  audio stream.

11
Encoding Decoding

• The decoding and synthesis process includes
  several steps.
• Decode packet type flag First the decoder must
  verify that a given packet contains audio data by
  inspecting its type flag.
• Decode mode number The mode number indicates
  the current frame size, window type, transform
  type and mapping number.
• The frame size is a power of 2 between 64 and
  8192, and can be either short or long. Short
  windows are used near attack transients in order
  to limit artifacts associated with the MDCT.
• The window taper varies for long windows
  depending on whether the previous and subsequent
  frames are short or long.
• The transform type is always type 0, the MDCT,
  in Vorbis I.
• The mapping number contains a description of the
  channel coupling scheme and a list of sub-maps
  which bundle sets of channel vectors.

12
Encoding Decoding

• 3. Decode window shape The window shape for a
  given long frame is decoded.
• Decode the floor The floor vector is a
  low-resolution representation of the audio
  spectrum for the given channel in the current
  frame, generally used akin to a whitening
  filter. During encoding this data is extracted
  from the log spectrum of the audio stream using
  either floor configuration type 0 or 1.
• Type 0 uses Line Spectral Pair (LSP, also
  alternately known as Line Spectral Frequency or
  LSF) representation to encode a smooth spectral
  envelope curve as the frequency response of the
  LSP filter. This representation is equivalent to
  a traditional all-pole infinite impulse response
  filter as would be used in linear predictive
  coding LSP representation may be converted to
  LPC representation and vice-versa.

13
Encoding Decoding

• Type 1 uses a piecewise straight-line
  representation to encode a spectral envelope
  curve. The representation plots this curve
  mechanically on a linear frequency axis and a
  logarithmic (dB) amplitude axis. The integer
  plotting algorithm used is similar to Bresenham's
  algorithm.
• 5. Decode the residue The high frequency detail
  remaining after the floor has been subtracted
  from the audio spectrum (for a given channel and
  frame) during encoding comprises the residue.

14
Encoding Decoding

• 6. Inverse channel coupling of residue vectors
  The bit rate is lowered during encoding by
  eliminating redundancies between channels.
• Two mechanisms exist for channel coupling
• Channel interleaving via residue backend type 2
• Cartesian to square polar mapping.
• The inverse process is performed during decoding.
  
• For encoder quality settings equal or greater
  than six, channel coupling is loseless.

15
Encoding Decoding

• The floor curve is generated from the decoded
  floor data.
• The dot product of the floor and residue vectors
  is taken to produce an audio spectrum vector.
• The audio spectrum is converted back to the time
  domain via the inverse MDCT.
• 10. The result of the transform is overlapped
  and added together frame-by-frame to provide the
  new audio stream.