4. Audio

Audio

4.1 Encoding-Independent Rules

Since the ability to suppress silence is one of the primary motivations for using packets to transmit voice, the RTP header carries both a sequence number and a timestamp to allow a receiver to distinguish between lost packets and periods of time when no data was transmitted. Discontiguous transmission (silence suppression) MAY be used with any audio payload format. Receivers MUST assume that senders may suppress silence unless this is restricted by signaling specified elsewhere. (Even if the transmitter does not suppress silence, the receiver should be prepared to handle periods when no data is present since packets may be lost.)

Some payload formats (see Sections 4.5.3 and 4.5.6) define a "silence insertion descriptor" or "comfort noise" frame to specify parameters for artificial noise that may be generated during a period of silence to approximate the background noise at the source. For other payload formats, a generic Comfort Noise (CN) payload format is specified in RFC 3389 [9]. When the CN payload format is used with another payload format, different values in the RTP payload type field distinguish comfort-noise packets from those of the selected payload format.

For applications which send either no packets or occasional comfort- noise packets during silence, the first packet of a talkspurt, that is, the first packet after a silence period during which packets have not been transmitted contiguously, SHOULD be distinguished by setting the marker bit in the RTP data header to one. The marker bit in all other packets is zero. The beginning of a talkspurt MAY be used to adjust the playout delay to reflect changing network delays. Applications without silence suppression MUST set the marker bit to zero.

The RTP clock rate used for generating the RTP timestamp is independent of the number of channels and the encoding; it usually equals the number of sampling periods per second. For N-channel encodings, each sampling period (say, 1/8,000 of a second) generates N samples. (This terminology is standard, but somewhat confusing, as the total number of samples generated per second is then the sampling rate times the channel count.)

If multiple audio channels are used, channels are numbered left-to- right, starting at one. In RTP audio packets, information from lower-numbered channels precedes that from higher-numbered channels.

For more than two channels, the convention followed by the AIFF-C audio interchange format SHOULD be followed [3], using the following notation, unless some other convention is specified for a particular encoding or payload format:

  l  left
  r  right
  c  center
  S  surround
  F  front
  R  rear

  channels  description  channel
                            1     2   3   4   5   6
  _________________________________________________
  2         stereo          l     r
  3                         l     r   c
  4                         l     c   r   S
  5                        Fl     Fr  Fc  Sl  Sr
  6                         l     lc  c   r   rc  S

     Note: RFC 1890 defined two conventions for the ordering of four
     audio channels.  Since the ordering is indicated implicitly by
     the number of channels, this was ambiguous.  In this revision,
     the order described as "quadrophonic" has been eliminated to
     remove the ambiguity.  This choice was based on the observation
     that quadrophonic consumer audio format did not become popular
     whereas surround-sound subsequently has.

Samples for all channels belonging to a single sampling instant MUST be within the same packet. The interleaving of samples from different channels depends on the encoding. General guidelines are given in Section 4.3 and 4.4.

The sampling frequency SHOULD be drawn from the set: 8,000, 11,025, 16,000, 22,050, 24,000, 32,000, 44,100 and 48,000 Hz. (Older Apple Macintosh computers had a native sample rate of 22,254.54 Hz, which can be converted to 22,050 with acceptable quality by dropping 4 samples in a 20 ms frame.) However, most audio encodings are defined for a more restricted set of sampling frequencies. Receivers SHOULD be prepared to accept multi-channel audio, but MAY choose to only play a single channel.

4.2 Operating Recommendations

The following recommendations are default operating parameters. Applications SHOULD be prepared to handle other values. The ranges given are meant to give guidance to application writers, allowing a set of applications conforming to these guidelines to interoperate without additional negotiation. These guidelines are not intended to restrict operating parameters for applications that can negotiate a set of interoperable parameters, e.g., through a conference control protocol.

For packetized audio, the default packetization interval SHOULD have a duration of 20 ms or one frame, whichever is longer, unless otherwise noted in Table 1 (column "ms/packet"). The packetization interval determines the minimum end-to-end delay; longer packets introduce less header overhead but higher delay and make packet loss more noticeable. For non-interactive applications such as lectures or for links with severe bandwidth constraints, a higher packetization delay MAY be used. A receiver SHOULD accept packets representing between 0 and 200 ms of audio data. (For framed audio encodings, a receiver SHOULD accept packets with a number of frames equal to 200 ms divided by the frame duration, rounded up.) This restriction allows reasonable buffer sizing for the receiver.

4.3 Guidelines for Sample-Based Audio Encodings

In sample-based encodings, each audio sample is represented by a fixed number of bits. Within the compressed audio data, codes for individual samples may span octet boundaries. An RTP audio packet may contain any number of audio samples, subject to the constraint that the number of bits per sample times the number of samples per packet yields an integral octet count. Fractional encodings produce less than one octet per sample.

The duration of an audio packet is determined by the number of samples in the packet.

For sample-based encodings producing one or more octets per sample, samples from different channels sampled at the same sampling instant SHOULD be packed in consecutive octets. For example, for a two- channel encoding, the octet sequence is (left channel, first sample), (right channel, first sample), (left channel, second sample), (right channel, second sample), .... For multi-octet encodings, octets SHOULD be transmitted in network byte order (i.e., most significant octet first).

The packing of sample-based encodings producing less than one octet per sample is encoding-specific.

The RTP timestamp reflects the instant at which the first sample in the packet was sampled, that is, the oldest information in the packet.

4.4 Guidelines for Frame-Based Audio Encodings

Frame-based encodings encode a fixed-length block of audio into another block of compressed data, typically also of fixed length. For frame-based encodings, the sender MAY choose to combine several such frames into a single RTP packet. The receiver can tell the number of frames contained in an RTP packet, if all the frames have the same length, by dividing the RTP payload length by the audio frame size which is defined as part of the encoding. This does not work when carrying frames of different sizes unless the frame sizes are relatively prime. If not, the frames MUST indicate their size.

For frame-based codecs, the channel order is defined for the whole block. That is, for two-channel audio, right and left samples SHOULD be coded independently, with the encoded frame for the left channel preceding that for the right channel.

All frame-oriented audio codecs SHOULD be able to encode and decode several consecutive frames within a single packet. Since the frame size for the frame-oriented codecs is given, there is no need to use a separate designation for the same encoding, but with different number of frames per packet.

RTP packets SHALL contain a whole number of frames, with frames inserted according to age within a packet, so that the oldest frame (to be played first) occurs immediately after the RTP packet header. The RTP timestamp reflects the instant at which the first sample in the first frame was sampled, that is, the oldest information in the packet.

4.5 Audio Encodings

name of sampling default encoding sample/frame bits/sample rate ms/frame ms/packet

DVI4 sample 4 var. 20 G722 sample 8 16,000 20 G723 frame N/A 8,000 30 30 G726-40 sample 5 8,000 20 G726-32 sample 4 8,000 20 G726-24 sample 3 8,000 20 G726-16 sample 2 8,000 20 G728 frame N/A 8,000 2.5 20 G729 frame N/A 8,000 10 20 G729D frame N/A 8,000 10 20 G729E frame N/A 8,000 10 20 GSM frame N/A 8,000 20 20 GSM-EFR frame N/A 8,000 20 20 L8 sample 8 var. 20 L16 sample 16 var. 20 LPC frame N/A 8,000 20 20 MPA frame N/A var. var. PCMA sample 8 var. 20 PCMU sample 8 var. 20 QCELP frame N/A 8,000 20 20 VDVI sample var. var. 20

Table 1: Properties of Audio Encodings (N/A: not applicable; var.: variable)

The characteristics of the audio encodings described in this document are shown in Table 1; they are listed in order of their payload type in Table 4. While most audio codecs are only specified for a fixed sampling rate, some sample-based algorithms (indicated by an entry of "var." in the sampling rate column of Table 1) may be used with different sampling rates, resulting in different coded bit rates. When used with a sampling rate other than that for which a static payload type is defined, non-RTP means beyond the scope of this memo MUST be used to define a dynamic payload type and MUST indicate the selected RTP timestamp clock rate, which is usually the same as the sampling rate for audio.

4.5.1 DVI4

DVI4 uses an adaptive delta pulse code modulation (ADPCM) encoding scheme that was specified by the Interactive Multimedia Association (IMA) as the "IMA ADPCM wave type". However, the encoding defined here as DVI4 differs in three respects from the IMA specification:

o The RTP DVI4 header contains the predicted value rather than the first sample value contained the IMA ADPCM block header.

o IMA ADPCM blocks contain an odd number of samples, since the first sample of a block is contained just in the header (uncompressed), followed by an even number of compressed samples. DVI4 has an even number of compressed samples only, using the `predict' word from the header to decode the first sample.

o For DVI4, the 4-bit samples are packed with the first sample in the four most significant bits and the second sample in the four least significant bits. In the IMA ADPCM codec, the samples are packed in the opposite order.

Each packet contains a single DVI block. This profile only defines the 4-bit-per-sample version, while IMA also specified a 3-bit-per- sample encoding.

The "header" word for each channel has the following structure:

  int16  predict;  /* predicted value of first sample
                      from the previous block (L16 format) */
  u_int8 index;    /* current index into stepsize table */
  u_int8 reserved; /* set to zero by sender, ignored by receiver */

Each octet following the header contains two 4-bit samples, thus the number of samples per packet MUST be even because there is no means to indicate a partially filled last octet.

Packing of samples for multiple channels is for further study.

The IMA ADPCM algorithm was described in the document IMA Recommended Practices for Enhancing Digital Audio Compatibility in Multimedia Systems (version 3.0). However, the Interactive Multimedia Association ceased operations in 1997. Resources for an archived copy of that document and a software implementation of the RTP DVI4 encoding are listed in Section 13.

4.5.2 G722

G722 is specified in ITU-T Recommendation G.722, "7 kHz audio-coding within 64 kbit/s". The G.722 encoder produces a stream of octets, each of which SHALL be octet-aligned in an RTP packet. The first bit transmitted in the G.722 octet, which is the most significant bit of the higher sub-band sample, SHALL correspond to the most significant bit of the octet in the RTP packet.

Even though the actual sampling rate for G.722 audio is 16,000 Hz, the RTP clock rate for the G722 payload format is 8,000 Hz because that value was erroneously assigned in RFC 1890 and must remain unchanged for backward compatibility. The octet rate or sample-pair rate is 8,000 Hz.

4.5.3 G723

G723 is specified in ITU Recommendation G.723.1, "Dual-rate speech coder for multimedia communications transmitting at 5.3 and 6.3 kbit/s". The G.723.1 5.3/6.3 kbit/s codec was defined by the ITU-T as a mandatory codec for ITU-T H.324 GSTN videophone terminal applications. The algorithm has a floating point specification in Annex B to G.723.1, a silence compression algorithm in Annex A to G.723.1 and a scalable channel coding scheme for wireless applications in G.723.1 Annex C.

This Recommendation specifies a coded representation that can be used for compressing the speech signal component of multi-media services at a very low bit rate. Audio is encoded in 30 ms frames, with an additional delay of 7.5 ms due to look-ahead. A G.723.1 frame can be one of three sizes: 24 octets (6.3 kb/s frame), 20 octets (5.3 kb/s frame), or 4 octets. These 4-octet frames are called SID frames (Silence Insertion Descriptor) and are used to specify comfort noise parameters. There is no restriction on how 4, 20, and 24 octet frames are intermixed. The least significant two bits of the first octet in the frame determine the frame size and codec type:

     bits  content                      octets/frame
  high-rate speech (6.3 kb/s)            24
  low-rate speech  (5.3 kb/s)            20
  SID frame                               4
  reserved

It is possible to switch between the two rates at any 30 ms frame boundary. Both (5.3 kb/s and 6.3 kb/s) rates are a mandatory part of the encoder and decoder. Receivers MUST accept both data rates and MUST accept SID frames unless restriction of these capabilities has been signaled. The MIME registration for G723 in RFC 3555 [7] specifies parameters that MAY be used with MIME or SDP to restrict to a single data rate or to restrict the use of SID frames. This coder was optimized to represent speech with near-toll quality at the above rates using a limited amount of complexity.

The packing of the encoded bit stream into octets and the transmission order of the octets is specified in Rec. G.723.1 and is the same as that produced by the G.723 C code reference implementation. For the 6.3 kb/s data rate, this packing is illustrated as follows, where the header (HDR) bits are always "0 0" as shown in Fig. 1 to indicate operation at 6.3 kb/s, and the Z bit is always set to zero. The diagrams show the bit packing in "network byte order", also known as big-endian order. The bits of each 32-bit word are numbered 0 to 31, with the most significant bit on the left and numbered 0. The octets (bytes) of each word are transmitted most significant octet first. The bits of each data field are numbered in the order of the bit stream representation of the encoding (least significant bit first). The vertical bars indicate the boundaries between field fragments.

0                   1                   2                   3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1

+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | LPC |HDR| LPC | LPC | ACL0 |LPC| | | | | | | | |0 0 0 0 0 0|0 0|1 1 1 1 0 0 0 0|2 2 1 1 1 1 1 1|0 0 0 0 0 0|2 2| |5 4 3 2 1 0| |3 2 1 0 9 8 7 6|1 0 9 8 7 6 5 4|5 4 3 2 1 0|3 2| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | ACL2 |ACL|A| GAIN0 |ACL|ACL| GAIN0 | GAIN1 | | | 1 |C| | 3 | 2 | | | |0 0 0 0 0|0 0|0|0 0 0 0|0 0|0 0|1 1 0 0 0 0 0 0|0 0 0 0 0 0 0 0| |4 3 2 1 0|1 0|6|3 2 1 0|1 0|6 5|1 0 9 8 7 6 5 4|7 6 5 4 3 2 1 0| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | GAIN2 | GAIN1 | GAIN2 | GAIN3 | GRID | GAIN3 | | | | | | | | |0 0 0 0|1 1 0 0|1 1 0 0 0 0 0 0|0 0 0 0 0 0 0 0|0 0 0 0|1 1 0 0| |3 2 1 0|1 0 9 8|1 0 9 8 7 6 5 4|7 6 5 4 3 2 1 0|3 2 1 0|1 0 9 8| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | MSBPOS |Z|POS| MSBPOS | POS0 |POS| POS0 | | | | 0 | | | 1 | | |0 0 0 0 0 0 0|0|0 0|1 1 1 0 0 0|0 0 0 0 0 0 0 0|0 0|1 1 1 1 1 1| |6 5 4 3 2 1 0| |1 0|2 1 0 9 8 7|9 8 7 6 5 4 3 2|1 0|5 4 3 2 1 0| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | POS1 | POS2 | POS1 | POS2 | POS3 | POS2 | | | | | | | | |0 0 0 0 0 0 0 0|0 0 0 0|1 1 1 1|1 1 0 0 0 0 0 0|0 0 0 0|1 1 1 1| |9 8 7 6 5 4 3 2|3 2 1 0|3 2 1 0|1 0 9 8 7 6 5 4|3 2 1 0|5 4 3 2| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | POS3 | PSIG0 |POS|PSIG2| PSIG1 | PSIG3 |PSIG2| | | | 3 | | | | | |1 1 0 0 0 0 0 0|0 0 0 0 0 0|1 1|0 0 0|0 0 0 0 0|0 0 0 0 0|0 0 0| |1 0 9 8 7 6 5 4|5 4 3 2 1 0|3 2|2 1 0|4 3 2 1 0|4 3 2 1 0|5 4 3| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

              Figure 1: G.723 (6.3 kb/s) bit packing

For the 5.3 kb/s data rate, the header (HDR) bits are always "0 1", as shown in Fig. 2, to indicate operation at 5.3 kb/s.

0                   1                   2                   3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1

+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | LPC |HDR| LPC | LPC | ACL0 |LPC| | | | | | | | |0 0 0 0 0 0|0 1|1 1 1 1 0 0 0 0|2 2 1 1 1 1 1 1|0 0 0 0 0 0|2 2| |5 4 3 2 1 0| |3 2 1 0 9 8 7 6|1 0 9 8 7 6 5 4|5 4 3 2 1 0|3 2| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | ACL2 |ACL|A| GAIN0 |ACL|ACL| GAIN0 | GAIN1 | | | 1 |C| | 3 | 2 | | | |0 0 0 0 0|0 0|0|0 0 0 0|0 0|0 0|1 1 0 0 0 0 0 0|0 0 0 0 0 0 0 0| |4 3 2 1 0|1 0|6|3 2 1 0|1 0|6 5|1 0 9 8 7 6 5 4|7 6 5 4 3 2 1 0| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | GAIN2 | GAIN1 | GAIN2 | GAIN3 | GRID | GAIN3 | | | | | | | | |0 0 0 0|1 1 0 0|1 1 0 0 0 0 0 0|0 0 0 0 0 0 0 0|0 0 0 0|1 1 0 0| |3 2 1 0|1 0 9 8|1 0 9 8 7 6 5 4|7 6 5 4 3 2 1 0|4 3 2 1|1 0 9 8| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | POS0 | POS1 | POS0 | POS1 | POS2 | | | | | | | |0 0 0 0 0 0 0 0|0 0 0 0|1 1 0 0|1 1 0 0 0 0 0 0|0 0 0 0 0 0 0 0| |7 6 5 4 3 2 1 0|3 2 1 0|1 0 9 8|1 0 9 8 7 6 5 4|7 6 5 4 3 2 1 0| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | POS3 | POS2 | POS3 | PSIG1 | PSIG0 | PSIG3 | PSIG2 | | | | | | | | | |0 0 0 0|1 1 0 0|1 1 0 0 0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0| |3 2 1 0|1 0 9 8|1 0 9 8 7 6 5 4|3 2 1 0|3 2 1 0|3 2 1 0|3 2 1 0| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

              Figure 2: G.723 (5.3 kb/s) bit packing

The packing of G.723.1 SID (silence) frames, which are indicated by the header (HDR) bits having the pattern "1 0", is depicted in Fig. 3.

0                   1                   2                   3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1

+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | LPC |HDR| LPC | LPC | GAIN |LPC| | | | | | | | |0 0 0 0 0 0|1 0|1 1 1 1 0 0 0 0|2 2 1 1 1 1 1 1|0 0 0 0 0 0|2 2| |5 4 3 2 1 0| |3 2 1 0 9 8 7 6|1 0 9 8 7 6 5 4|5 4 3 2 1 0|3 2| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

               Figure 3: G.723 SID mode bit packing

4.5.4 G726-40, G726-32, G726-24, and G726-16

ITU-T Recommendation G.726 describes, among others, the algorithm recommended for conversion of a single 64 kbit/s A-law or mu-law PCM channel encoded at 8,000 samples/sec to and from a 40, 32, 24, or 16 kbit/s channel. The conversion is applied to the PCM stream using an Adaptive Differential Pulse Code Modulation (ADPCM) transcoding technique. The ADPCM representation consists of a series of codewords with a one-to-one correspondence to the samples in the PCM stream. The G726 data rates of 40, 32, 24, and 16 kbit/s have codewords of 5, 4, 3, and 2 bits, respectively.

The 16 and 24 kbit/s encodings do not provide toll quality speech. They are designed for used in overloaded Digital Circuit Multiplication Equipment (DCME). ITU-T G.726 recommends that the 16 and 24 kbit/s encodings should be alternated with higher data rate encodings to provide an average sample size of between 3.5 and 3.7 bits per sample.

The encodings of G.726 are here denoted as G726-40, G726-32, G726-24, and G726-16. Prior to 1990, G721 described the 32 kbit/s ADPCM encoding, and G723 described the 40, 32, and 16 kbit/s encodings. Thus, G726-32 designates the same algorithm as G721 in RFC 1890.

A stream of G726 codewords contains no information on the encoding being used, therefore transitions between G726 encoding types are not permitted within a sequence of packed codewords. Applications MUST determine the encoding type of packed codewords from the RTP payload identifier.

No payload-specific header information SHALL be included as part of the audio data. A stream of G726 codewords MUST be packed into octets as follows: the first codeword is placed into the first octet such that the least significant bit of the codeword aligns with the least significant bit in the octet, the second codeword is then packed so that its least significant bit coincides with the least significant unoccupied bit in the octet. When a complete codeword cannot be placed into an octet, the bits overlapping the octet boundary are placed into the least significant bits of the next octet. Packing MUST end with a completely packed final octet. The number of codewords packed will therefore be a multiple of 8, 2, 8, and 4 for G726-40, G726-32, G726-24, and G726-16, respectively. An example of the packing scheme for G726-32 codewords is as shown, where bit 7 is the least significant bit of the first octet, and bit A3 is the least significant bit of the first codeword:

      0                   1
      0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
     |B B B B|A A A A|D D D D|C C C C| ...
     |0 1 2 3|0 1 2 3|0 1 2 3|0 1 2 3|
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-

An example of the packing scheme for G726-24 codewords follows, where again bit 7 is the least significant bit of the first octet, and bit A2 is the least significant bit of the first codeword:

      0                   1                   2
      0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
     |C C|B B B|A A A|F|E E E|D D D|C|H H H|G G G|F F| ...
     |1 2|0 1 2|0 1 2|2|0 1 2|0 1 2|0|0 1 2|0 1 2|0 1|
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-

Note that the "little-endian" direction in which samples are packed into octets in the G726-16, -24, -32 and -40 payload formats specified here is consistent with ITU-T Recommendation X.420, but is the opposite of what is specified in ITU-T Recommendation I.366.2 Annex E for ATM AAL2 transport. A second set of RTP payload formats matching the packetization of I.366.2 Annex E and identified by MIME subtypes AAL2-G726-16, -24, -32 and -40 will be specified in a separate document.

4.5.5 G728

G728 is specified in ITU-T Recommendation G.728, "Coding of speech at 16 kbit/s using low-delay code excited linear prediction".

A G.278 encoder translates 5 consecutive audio samples into a 10-bit codebook index, resulting in a bit rate of 16 kb/s for audio sampled at 8,000 samples per second. The group of five consecutive samples is called a vector. Four consecutive vectors, labeled V1 to V4 (where V1 is to be played first by the receiver), build one G.728 frame. The four vectors of 40 bits are packed into 5 octets, labeled B1 through B5. B1 SHALL be placed first in the RTP packet.

Referring to the figure below, the principle for bit order is "maintenance of bit significance". Bits from an older vector are more significant than bits from newer vectors. The MSB of the frame goes to the MSB of B1 and the LSB of the frame goes to LSB of B5.

               1         2         3        3
     0         0         0         0        9
     ++++++++++++++++++++++++++++++++++++++++
     <---V1---><---V2---><---V3---><---V4---> vectors
     <--B1--><--B2--><--B3--><--B4--><--B5--> octets
     <------------- frame 1 ---------------->

In particular, B1 contains the eight most significant bits of V1, with the MSB of V1 being the MSB of B1. B2 contains the two least significant bits of V1, the more significant of the two in its MSB, and the six most significant bits of V2. B1 SHALL be placed first in the RTP packet and B5 last.

4.5.6 G729

G729 is specified in ITU-T Recommendation G.729, "Coding of speech at 8 kbit/s using conjugate structure algebraic-code-excited linear- prediction (CS-ACELP)". This codec was optimized to represent speech with high quality, where G.728 is low delay and G.723.1 is low bitrate. The G.729 encoder operates on speech frames of 10 ms corresponding to 80 samples at a sampling rate of 8000 samples per second. The generated bit stream has a bit rate of 8 kbit/s. Therefore, each frame contains 80 bits.

The packing of the encoded bit stream into octets and the transmission order of the octets is specified in Rec. G.729 and is the same as that produced by the G.729 C code reference implementation. The bits of each 80-bit frame are numbered 1 to 80, with the most significant bit on the left and numbered 1. The octets (bytes) of each word are transmitted most significant octet first. The bits of each data field are numbered in the order of the bit stream representation of the encoding (least significant bit first). The vertical bars indicate the boundaries between field fragments.

0                   1                   2                   3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1

+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |L0 | L1 | L2 | L3 | P1 | | | | | | | |0|0 0 0 0 0 0 0|0 0 0 0 0|0 0 0 0 0|0 0 0 0 0 0 0 0| | |6 5 4 3 2 1 0|4 3 2 1 0|4 3 2 1 0|7 6 5 4 3 2 1 0| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P1 | P2 | C1 | S1 | C2 | S2 | P1 | | | | | | | | | |0 0|0 0 0|0 0 0 0 0|0 0 0|0 0 0 0 0|0 0 0|0 0 0 0 0 0 0 0| |9 8|4 3 2 1 0|c c c c c|3 2 1 0|c c c c c|3 2 1 0|7 6 5 4 3 2 1 0| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | P2 | C3 | S3 | C4 | S4 | | | | | | | |0 0 0|0 0 0 0 0|0 0 0|0 0 0 0 0|0 0 0| |4 3 2 1 0|c c c c c|3 2 1 0|c c c c c|3 2 1 0| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Annex A to G.729 specifies a reduced complexity version of the G.729 algorithm. The Annex A coder produces a bit stream identical to that of the G.729 coder.

Annex B to G.729 defines a silence compression algorithm, which is used with G.729 or G.729 Annex A. Detecting a silence period, the encoder transmits a 2-octet SID frame (Silence Insertion Descriptor) which specifies comfort noise parameters. There is no restriction on how 10 and 2 octet frames are intermixed. The SID frame is partially sum codebook indices and partially parameters of the silence compression algorithm. The fields of the SID frame are shown in the diagram below.

0                   1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5

+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |L0 | L1 | L2 | L3 | GAIN | | | | | | | |0|0 0 0 0 0 0 0|0 0 0 0 0|0 0 0 0 0|0 0 0 0 0| | |6 5 4 3 2 1 0|4 3 2 1 0|4 3 2 1 0|4 3 2 1 0| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

For G.729 Annex B, the channel-coded VAD/DTX generation includes the transmission of 1/16th rate (800 bps) frames that are just 10 bits in length (1.25 octets). Since these do not fit into an integral number of octets, the packing of these frames remains for future study. Receivers operating under this payload format MUST accept the 80-bit speech frames and the 16-bit SID frames. The receiver MAY discard 10-bit frames, if any.

4.5.7 G729D and G729E

Annex D to G.729 defines a lower bit rate extension (6.4 kb/s) to the algorithm. The G.729 Annex D coder produces a 64-bit frame.

Annex E to G.729 defines a higher bit rate extension (11.8 kb/s) to the algorithm. The G.729 Annex E coder produces a 118-bit frame.

The packing of the encoded bit stream into octets and the transmission order of the octets is specified in Rec. G.729 and is the same as that produced by the G.729 C code reference implementation. The bits of each 64-bit or 118-bit frame are numbered 1 to n, with the most significant bit on the left and numbered 1. The octets (bytes) of each word are transmitted most significant octet first. The bits of each data field are numbered in the order of the bit stream representation of the encoding (least significant bit first).

ITU-T Rec. G.729 Annexes D and E do not currently specify how Voice Activity Detection (VAD) and Discontinuous Transmission (DTX) should be handled. Therefore, they SHOULD NOT be used with this payload format.

4.5.8 GSM

GSM (Group Speciale Mobile) denotes the European GSM 06.10 standard for full-rate speech transcoding, ETS 300 961, which is based on RPE/ LTP (residual pulse excitation/long term prediction) coding at a rate of 13 kb/s [11,12,13]. The text of the standard can be obtained from the ETSI (European Telecommunications Standards Institute) web site at http://www.etsi.org.

4.5.8.1 General Packaging Issues

The GSM standard specifies the bit stream produced by the codec, but does not specify how these bits should be packed for transmission. Unlike the payload formats for the other audio encodings specified in this document, the GSM payload format packs the bytes differently from the standard representation used by the GSM 06.10 C reference implementation.

The standard GSM 06.10 implementation packages the 260 bits of one GSM frame into 33 octets (16 words in 16-bit systems), with the bits occupying the least significant bits of each byte/word. To transmit these in RTP, the 260 bits are packed into 33 octets as defined in ETSI TS 101 318 [4]. The latter is effectively being standard for VoIP/VoFR/VoATM and is the same as the "Toast" format used by the GSM implementation cited in Section 13.

In the RTP payload format, a frame is packed into 33 octets (264 bits) by padding the 4 spare bits of the last octet with zeros. Two frames are packed into 66 octets. The field mapping is shown in Table 2.

4.5.8.2 GSM Variable Names and Numbers

The parameters of the GSM audio codec are named in the following manner, according to the GSM 06.10 standard.

                         Table 2: GSM payload format

Field Field Name Bits Octet Bit

1 LARc[0] 6 1 0--5 2 LARc[1] 6 1, 2 6--7, 0--3 3 LARc[2] 5 2, 3 4--7, 0 4 LARc[3] 5 3 1--5 5 LARc[4] 4 3, 4 6--7, 0--1 6 LARc[5] 4 4 2--5 7 LARc[6] 3 4, 5 6--7, 0 8 LARc[7] 3 5 1--3 9 Nc[0] 7 5, 6 4--7, 0--2 10 bc[0] 2 6 3--4 11 Mc[0] 2 6 5--6 12 xmaxc[0] 6 6, 7 7, 0--4 13 xmc[0] 3 7 5--7 14 Nc[1] 7 8 0--6 15 bc[1] 2 8, 9 7, 0 16 Mc[1] 2 9 1--2 17 xmaxc[1] 6 9, 10 3--7, 0 18 xmc[1] 3 10 1--3 19 Nc[2] 7 10, 11 4--7, 0--2 20 bc[2] 2 11 3--4 21 Mc[2] 2 11 5--6 22 xmaxc[2] 6 11, 12 7, 0--4 23 xmc[2] 3 12 5--7 24 Nc[3] 7 13 0--6 25 bc[3] 2 13, 14 7, 0 26 Mc[3] 2 14 1--2 27 xmaxc[3] 6 14, 15 3--7, 0 28 xmc[3] 3 15 1--3

4.5.9 GSM-EFR

GSM-EFR is specified in ETS 300 726 (GSM 06.60). This is an enhanced full rate speech codec, with the same clock rate (8000 Hz) and bit rate (12.2 kb/s) as the GSM codec.

4.5.10 L8

L8 denotes linear audio data samples, using 8-bits of precision with an offset of 128, that is, the most negative signal is represented by the value 0, the zero signal by the value 128, and the most positive signal by the value 255.

4.5.11 L16

L16 denotes (16-bit) signed linear audio data samples. Linear L16 audio samples SHOULD be transmitted in network byte order (most significant octet first).

4.5.12 LPC

LPC designates an experimental linear predictive encoding contributed by Ron Frederick at Xerox PARC, which is based on an implementation originally written by Steve Casner at ISI. The source code for the encoder and decoder is available as listed in Section 13.

4.5.13 MPA

MPA designates the use of MPEG-1 or MPEG-2 audio elementary streams. The RTP payload format is as specified in RFC 2250 [14], Section 3.

4.5.14 PCMA and PCMU

PCMA and PCMU are specified in ITU-T Recommendation G.711. Audio data is encoded as eight bits per sample, after logarithmic scaling. PCMU denotes mu-law scaling, PCMA A-law scaling. A detailed description is given by Jayant and Noll [15]. Each G.711 octet SHALL be octet-aligned in an RTP packet. The sign bit of each G.711 octet SHALL correspond to the most significant bit of the octet in the RTP packet (i.e., assuming the G.711 samples are handled as octets on the host machine, the sign bit shall be the most significant bit of the octet as defined by the host machine format). The 56 kb/s and 48 kb/s modes of G.711 are not applicable to RTP, since PCMA and PCMU MUST always be transmitted as 8-bit samples.

4.5.15 QCELP

The Qualcomm Code Excited Linear Prediction (QCELP) audio codec is specified in TIA/EIA IS-733, "TR45: High Rate Speech Service Option 17 for Wideband Spread Spectrum Communication Systems".

The QCELP codec is used for the TIA/EIA IS-95 CDMA cellular phone system standard. The QCELP codec scales to different data rates, which can be selected on a frame by frame basis. The maximum data rate is 13 kbit/s. This payload format defines no separate silent frame type? but the data rate can be reduced to approx. 1 kbit/s in silent periods.

The RTP payload format is specified in [16].

4.5.16 RED

The redundant audio payload format "RED" is specified in RFC 2198 [17]. It defines a means by which multiple redundant copies of an audio packet may be transmitted in a single RTP stream.

4.5.17 VDVI

VDVI is a variable-rate version of DVI4, available for dynamic assignment. DVI4 samples are packed into octets, but the number of bits per sample can vary from packet to packet and is specified in the packet header. For VDVI, the payload type field of the RTP header is a dynamic type mapping to "VDVI".

Each packet contains a single DVI block. The "header" word for each channel has the following structure:

  int16  predict;  /* predicted value of first sample
                      from the previous block (L16 format) */
  u_int8 index;    /* current index into stepsize table */
  u_int8 sample_size; /* number of bits per sample */

Sample-size MUST be one of 3, 4, or 5. If sample_size is 4, the packing of samples is the same as for DVI4. If sample_size is 3 or 5, the samples are packed into the next available bits of the current octet of the packet window, beginning with the MSB. When a sample straddles an octet boundary, the bits are placed in the LSBs of the first octet and the MSBs of the second octet. The next sample is then packed starting at the first available MSB of the second octet. The process continues for the entire packet.