|Sampling frequency||300Hz to 3,4 kHz|
G.729 is a royalty-free audio codec (actually vocoder, voice coder, see Parametric Audio Coding) for compressing speech into digital signals. The technical name is "Conjugate Structure Algebraic Code Excited Linear Prediction" (CS-ACELP). G.729 Annexes A and B are used in Internet telephony , for example, due to their high compression and low computational effort.
G.729 is a hybrid compression method based on the investigation and transmission of speech parameters, with a so-called vocoder, as well as differential information and subsequent speech synthesis. The codec splits the audio signal into frames of 10 milliseconds length, which it examines for typical speech characteristics. These are set in parameters for later synthesis. In addition, the codec transmits differential information resulting from the artificially generated signal and the actual signal. Typically, two frames of 10 milliseconds each are transmitted together in a language packet, resulting in a delay of approximately 25 milliseconds.
This codec can only process audio signals that do not represent human speech as a source with difficulty. For example, it cannot process the multifrequency sounds used in analog telephony adequately. Here you can make do by filtering the multifrequency tones out of the signal and transmitting them according to RFC 2833 in the information channel ("outband").
G.729 also suppresses speech pauses. To prevent this from sounding like a disconnection to the listener, the decoder has the ability to fill speech pauses with so-called comfort noise. The standard includes possible implementations in both fixed-point and technically more complex floating point formats, which facilitates use in various complex DSP platforms. For these reasons, depending on the variant used, G.729 is comparatively computationally complex; depending on the implementation and the options it contains, it requires about 50 MIPS. The G.729A and G.729B variants have a low computing complexity and, for example, require around 10.3 million clock cycles for 80 audio samples in the non-optimized reference implementation of the ITU-T on the MicroBlaze microcontroller. However, depending on the architecture and type of optimization used, the MIPS specifications may deviate from the specified values and represent only rough guidelines.
G.729 is divided into different variants, in the standard as Annexes. These appendices are marked with different letters and other symbols for differentiation. Each appendix describes different possible combinations, which differ in the implementation effort, the required computing power and the functional scope of the codec. For correct decoding, the encoder and decoder must be matched to each other.
The following variants are available within G.729:
|Data rate 8 kbit/s||X||X||X||X||X||X||X||X||X||X||X||X|
|Data rate 6.4 kbit/s||X||X||X||X||X|
|Data rate 11.8 kbit/s||X||X||X||X||X|
The acronym DTX stands for discontinuous transmission, where on the transmitter side speech pauses, in which actually only contentless noise would have to be transmitted, are detected and transmitted in the form of bandwidth-saving pause signals, which are reproduced on the receiver side as locally generated comfort noise. In the Mean Opinion Score (MOS) G.729 achieves a perceived quality of 3.98 out of 5 points, whereas the variant G.729A achieves only 3.7 out of 5 points.
The most commonly used variants of the codec are Annex A and B, which use a fixed bit rate of 8 kbit/s for the coded voice signal, but in some variants fixed bit rates of 6.4 kbit/s and 11.8 kbit/s are also possible. The frequency spectrum ranges from 300 to 3400 Hz, whereby only voice data are transmitted accurately due to the coding concept.
The latest extension G.729J - this variant corresponds to the working designation G.729.1 - has the capability for broadband speech and audio coding: The transmitted frequency bandwidth was increased to the range 50 Hz to 7 kHz. The G.729J codec is hierarchically organized and the concrete bit rate and thus also the voice/audio quality can be set to variable bit rates by simply "cutting" the bit stream.
The mentioned data rate of 8 kbit/s is nominal, it refers exclusively to the audio data itself. If a data stream is now sent through a network, the overhead of the switching data for the data packets in which the data stream is packed is added. When using RTP in an IPv4 network this is 40 bytes per IPv4 data packet (60 bytes with IPv6). The frame length at G.729 is 10 ms and such a frame is encoded with 10 bytes. Typically 2 frames are sent per IPv4 data packet. Consequently, this setting effectively requires 60 bytes (40 + 10 + 10 bytes) for 20 ms voice data. This is 3000 bytes per second, i.e. 24 kbit/s (3000 bytes * 8 / 1000 = 24 kbit). If you pack more than 2 frames into one packet, the relative share of the IP data decreases and the overhead becomes smaller. With 3 frames per packet you would only need 18.7 kbit/s. However, the disadvantage is a longer delay: If this is still 25 ms (10 ms per frame + 5 ms processing time) at 2 frames per packet, this is already 35 ms at three frames. If the delay becomes too great, users may find it disturbing.