Recommendation ITU-T G.728 contains the description of an algorithm for the coding of speech signals at 16 kbit/s using low-delay code excited linear prediction (LD-CELP). The main description is found in the main body of the Recommendation, where the floating-point implementation of LD-CELP is described. Annexes A, B, C and D contain tables of constants used by the LD-CELP algorithm. In Annex E, the sequencing of variable adaptation and its use is given and Annex F lists the abbreviations used in ITU-T G.728.

Annex G contains the alternative16 kbit/s (16-bit) fixed-point specification of ITU-T G.728. The purpose of this Annex is to describe in sufficient detail how ITU-T G.728 operating at 16 kbit/s can be implemented on a fixed-point arithmetic device. A fixed-point implementation based on this description should be capable of fully interworking with a floating-point version of ITU-T G.728 and producing an output signal of equivalent quality, whether that signal is speech or an in-band data signal.

ITU-T G.728 LD-CELP can operate at three bitrates other than 16 kbit/s, as defined in Annexes H and J. These additional operating modes are particularly useful for applications such as digital circuit multiplication equipment (DCME).

Annex H contains the modifications to ITU-T G.728 LD-CELP speech coding algorithm needed to reduce the coding bit rate down to 12.8 and 9.6 kbit/s. These modifications include the modifications to the shape and gain codebooks. Operation at lower rates can assist DCME in handling increased voice traffic while showing a graceful quality degradation.

Annex I is concerned with how to conceal the loss of bit-stream information due to frame erasure or packet loss in the communications channel. During normal operation, the decoder performs in a manner identical with ITU-T G.728 (main body) or its Annex G when operating at 16 kbit/s, or with ITU-T G.728 Annex H when operating at 12.8 or 9.6 kbit/s. The modification described in Annex I only involves changing the decoder during times when the bit-stream is unavailable. It is presumed that this loss of the bit stream is indicated to the decoder by some external means. Annex I contains (for historical reasons) its own Appendix I.I, which provides information on values used for scaling ETPAST (an array storing current and past gain-scaled excitation values) during frame erasures. Annex I is not essential for normal operation of ITU-T G.728.

Annex J defines a 40 kbit/s extension that is optimized for voiceband data (VBD). The algorithmic delay is five samples long (0.625 ms), which is the same as the other LD-CELP modes of operation described in this Recommendation. The 40 kbit/s VBD algorithm in Annex J is intended for VBD signal compression transmission in applications such as DCME. The algorithm allows soft transition to and from the other ITU-T G.728 LD-CELP operating modes, and is also designed to maintain toll-quality speech. Annex J is an implementation-efficient alternative to the 40 kbit/s ADPCM mode (ITU-T G.726) in DCME systems that incorporate LD-CELP (ITU-T G.728).

Appendix I contains information on the implementation verification procedures for LD-CELP operating at 16 kbit/s (both floating- and fixed-point operation). This Appendix describes the digital test sequences and the measurement software to be used for implementation verification of ITU-T G.728. Provision is included for both floating-point implementations, based on the ITU-T G.728 main body, and bit-exact fixed-point implementations, based on Annex G.

Appendix II gives a broad outline of the speech performance of the 16 kbit/s LD-CELP algorithm when interacting with other parts of the network. Some general guidance is also offered on voice-like and non-voice signals. The performance of ITU-T G.728 is compared to that of 32 kbit/s operating mode of the ITU-T G.726 ADPCM)and of ITU-T G.711 (64 kbit/s PCM).