HD voice

Don't let your spectrum get squeezed. Leverage HD codecs for improved Voice AI performance

Standard call quality (8 kHz) and HD voice (16 kHz)

Typical copper telephony (PSTN) works in 8 kHz mode (g711 codec like PCMU/PCMA encodes/decodes signal at 8 kHz sampling rate)

Modern cloud telephony and mobile phones can operate at 16 kHz resulting in much improved signal quality. Calls running at 16 kHz are referred to as HD voice

Standard call quality (8 kHz)

HD voice call quality (16 kHz)

When a call with HD voice travels through the internet, it is required for all the intermediate hops to support 16 kHz for the call to stay at the HD quality. However, support for 16 kHz is not ubiquitous yet amongst the core service providers and telecoms

If any service on a call path happens to miss a support for HD voice, or if call is sent through the old copper line, a quality of signal gets irreversibly reduced to the standard quality of 8 kHz telephony

Voice AI and HD codecs

It gets even more important in context of Voice AI

With HD codecs all the extra information conveyed in the 4 - 8 kHz frequency band (relatively high frequencies) is available to STT or STS functions and can make significant improvement in performance 🚀

💡 When an input signal is sampled at a given sampling rate, frequencies up to half that rate can be represented in the output (sampling signals at standard telephony quality of 8 kHz preserves frequencies up to 4 kHz, signal sampled at 16 kHz sampling rate will contain frequencies up to 8 kHz of the input [original signal])

8 kHz sampling. Signal energy is present in the 0 - 4 kHz range after 8 kHz sampling. There is no data in 4 - 8 kHz range

16 kHz sampling. HD voice allows to catch signal energy in a wide 0 - 8 kHz range. Extra information present in 4 - 8 kHz range (compared to standard call quality) improves performance of Voice AI

HD voice support on VoIP Number

VoIP Number has support for HD voice with codecs:

• OPUS

• G722

• AMR-WB

OPUS codec can operate at even better quality encoding audio at rates up to 48 kHz

Codec	Sampling rate [kHz]	Bitrate [kbit/s]	Quality
OPUS	8	all bitrates 6 - 510 kbit/s supported (8 - 12 kbit/s sufficient)	narrowband
OPUS	12	all bitrates 6 - 510 kbit/s supported (16 - 20 kbit/s sufficient)	medium-band
OPUS	16	all bitrates 6 - 510 kbit/s supported (28 - 40 kbit/s sufficient)	wideband (HD voice)
OPUS	24	all bitrates 6 - 510 kbit/s supported (48 - 64 kbit/s sufficient)	super-wideband
OPUS	48	all bitrates 6 - 510 kbit/s supported (64 - 128 kbit/s sufficient)	fullband
G722	16	64	wideband (HD voice)
AMR-WB	16	6.60, 8.85, 12.65, 14.25, 15.85, 18.25, 19.85, 23.05, 23.85	wideband (HD voice)