Voice

HD Voice

HD Voice is wideband telephone audio: roughly 50-7000 Hz instead of the classic 300-3400 Hz narrowband channel. Both ends of the call, and every hop between them, must carry a wideband codec without transcoding down to narrowband, or the call falls back to standard narrowband audio.

Updated August 27, 2026

The phrase HD Voice is a marketing label that stuck to a real engineering change: raising the audio bandwidth of a telephone call beyond the analog PSTN filter. Traditional telephony, standardized around G.711 and the analog local loop, passes about 300-3400 Hz. That range covers most vowel energy and intelligibility, which is why it was enough for a century of circuit-switched calling. It also cuts out consonant detail, music, and the spatial cues that make a voice sound present.

Wideband speech, as used in HD Voice, typically covers about 50-7000 Hz (ITU-T G.722's nominal band). Super-wideband and fullband go further (Opus can encode up to 20 kHz). On a good path the result is easier listening, better talker recognition, and a higher Mean Opinion Score. On a mixed path the call is only as wide as the narrowest hop.

What HD means in telephony

HD here is not a pixel count. It refers to the analog bandwidth of the recovered speech, which is a function of sample rate and codec. Narrowband telephony samples at 8 kHz (Nyquist limit 4 kHz, with a practical passband of about 300-3400 Hz after filtering). Wideband samples at 16 kHz. Super-wideband often uses 32 kHz. Fullband uses 48 kHz, the same family of rates as consumer audio.

GSMA and 3GPP popularized 'HD Voice' for mobile networks that deploy AMR-WB (Adaptive Multi-Rate Wideband, also published as ITU-T G.722.2). Fixed and VoIP systems more often advertise G.722 or Opus. The listener does not hear the brand. They hear whether energy above 3.4 kHz survived the path.

Narrowband versus wideband spectrum

Speech energy is not uniform. Fundamental pitch for adult talkers often sits between about 85 and 250 Hz. Fricatives (/s/, /f/, /th/) carry identification cues well above 4 kHz. Narrowband telephony therefore preserves vowels better than sibilants. Wideband restores those cues and the low-frequency warmth that analog loops also discarded. The perceptual gap is largest on speakerphone, in noise, and when the talker is not a familiar voice.

Passbands are nominal. Filters, transcoding, and acoustic devices can narrow them.
ProfileTypical sample rateApproximate passbandCommon codecsWhere it appears
Narrowband8 kHz300-3400 HzG.711 μ-law/A-law, G.729, AMR-NBClassic PSTN, default SIP trunk to the PSTN
Wideband (HD Voice)16 kHz50-7000 HzG.722, AMR-WB (G.722.2)VoLTE HD, many SIP phones, enterprise VoIP
Super-wideband / fullband32-48 kHzUp to ~20 kHzOpus, some EVS modesWebRTC, app-to-app, high-end conferencing

Codecs that carry HD Voice

Three families dominate production HD calling:

  • ITU-T G.722 (1988): 16 kHz SB-ADPCM at 48, 56, or 64 kbit/s. Widely implemented on SIP desk phones. Not the same algorithm as G.722.1 or G.722.2.
  • AMR-WB / G.722.2 (3GPP): nine modes from 6.60 to 23.85 kbit/s. The mobile HD Voice codec for UMTS and VoLTE when the radio and core allow it.
  • Opus (IETF RFC 6716): hybrid SILK/CELT, typically 6-510 kbit/s, up to fullband. Default audio codec in WebRTC. Royalty-free.

A call remains HD only as long as every transcoding point in the path keeps a wideband codec. An SBC that interworks G.722 to G.711 for a PSTN hop throws the extra band away. Two G.722 endpoints behind that SBC will still sound narrowband to each other if the media is anchored and transcoded in the middle.

Both legs must support it

Codecs are negotiated, not assumed. In SIP, each endpoint lists payload types in SDP (a=rtpmap). The intersection of the two lists, after any middlebox policy, is the operational codec. In WebRTC, the same offer/answer happens via JSEP. If one side only offers PCMU (G.711 μ-law), the call is narrowband regardless of the other side's Opus support.

This is why 'HD Voice' on a mobile plan does not guarantee HD on a call to a landline, a contact center, or an older PBX. The radio hop may be AMR-WB while the interconnect is G.711. The same rule applies inside a company: enable G.722 on the phones and on the trunk, or the phones will fall back as soon as a call leaves the LAN.

Measuring the improvement

Subjective quality is still scored with Mean Opinion Score methods from ITU-T P.800 (a 1-5 listening scale). Objective estimators such as POLQA (ITU-T P.863) replace older PESQ for wideband and super-wideband. Typical laboratory scores put G.711 around the low 4s and well-implemented G.722 or AMR-WB higher, with Opus fullband at the top of the telephone range when the network is clean.

MOS is not a bandwidth meter. A wideband call with 3% packet loss can score worse than a clean G.711 call. HD Voice pays off when latency, jitter, and loss are already under control. It does not fix a congested Wi-Fi hop.

HD Voice versus standard telephony

Standard (narrowband)HD Voice (wideband)
PassbandAbout 300-3400 HzAbout 50-7000 Hz (G.722 / AMR-WB)
Sample rate8 kHz16 kHz (higher for Opus fullband)
Typical codecsG.711, G.729, AMR-NBG.722, AMR-WB, Opus
IntelligibilityAdequate for quiet, known talkersBetter consonants, speaker identity, and noisy rooms
RequirementUniversal on the PSTNBoth ends and every hop between them must support a wideband codec
FallbackN/A (this is the fallback)G.711 or another narrowband codec when HD is unavailable

For implementation detail on bitrates, licenses, and which codec to offer first, see audio codecs. For the packet path that carries them, see VoIP.