G.722Also known as HD voice, Wideband audio, SB-ADPCM

How G.722 works

G.722 splits the audio into a low band (0–4 kHz) and a high band (4–8 kHz) with a quadrature mirror filter, then encodes each with ADPCM, spending most bits on the low band where speech energy is. The result keeps consonants such as “s”, “f” and “th” that narrowband G.711 cuts off, making speech more intelligible and less tiring, while using the same 64 kbit/s.

Algorithmic delay is very low and CPU cost is modest, which is why it is common on desk phones, softphones and embedded devices.

The SDP clock-rate quirk

For historical reasons, RFC 3551 registers G.722 with RTP payload type 9 and a clock rate of 8000 Hz, even though it samples at 16 kHz. SDP therefore says “G722/8000”. Implementations must follow this convention; getting it wrong breaks timestamps and jitter calculations.

HD voice end to end

Wideband only helps if every leg carries it. A single narrowband hop, such as a G.711 trunk, reduces the whole call to telephone quality. On mobile networks the equivalent is AMR-WB or EVS on VoLTE.

The GSMCalls Smart Bridge is a microcontroller link between the phone and the platform that carries wideband G.722 audio without a PC in the path; see GSM gateway without a PC. For how codecs compare, see codec and Opus.