Adaptive multi-rate

from Wikipedia, the free encyclopedia
Adaptive multi-rate
File extension : .awb
MIME type : audio / amr-wb, audio / 3gpp
Type: Audio format
Contained in: .3gp
Extended by: ACELP
Expanded to: AMR-WB +, VMR-WB
Standard (s) : ITU-T G.722.2 ,

3GPP TS 26.171 ( ZIP ; 193 kB)

Adaptive Multi-Rate ( AMR , AMR-NB ) is a specialized in voice signals on the transmission standard ACELP -based audio format to lossy audio data compression and substantially an extension of Enhanced Full Rate (EFR) for a further bit rate modes.

Adaptive Multi-Rate Wideband ( AMR-WB , TS 26.171 , G.722.2 ) essentially adds the ability to reproduce an additional octave of reproducible frequency range to the original AMR. Extended Adaptive Multi-Rate - Wideband ( AMR-WB + ) is a further development that uses Transform Coded Excitation (TCX) in addition to ACELP in order to be able to offer better performance with signals other than voice, as well as stereo signals and even higher sampling frequencies enables. The Enhanced Voice Service ( EVS ) from LTE Rel-10 can supplement a conventional AMR-WB layer with additional data layers, with which mainly an extension of the reproducible frequency range is to be achieved, but also multi-channel sound is to be possible.

The company Ericsson , Nokia and VoiceAge Corporation ( Université de Sherbrooke ) keep software patents in ACELP. VoiceAge charges license fees for the use of the entire patent pool.

Features and technology

The entire format family is essentially based on Algebraic Code Excited Linear Prediction (ACELP).

Adaptive Multi-Rate Wideband (AMR-WB)

AMR-WB was developed for broadband voice connections in mobile radio systems of the second and third generation and is an international standard under the designations "TS 26.171" (ETSI / 3GPP) and "G.722.2" (ITU-T).

With its relatively high sampling rate of 16 kHz, it is a so-called "broadband" format. With the use of AMR-WB, the frequency range that can be transmitted is expanded by one octave compared to the usual telephone quality of 3.4 kHz to around 6.4 kHz or 7 kHz. The method should also be able to better transmit mixed signals with speech and ambient noises and enable better speech quality even in a noisy environment.

Several variants are defined, each with a different data rate of between 1.75 (Silence Descriptor, SID, for silence) or 6.6 and 23.85 kBit per second. The main variant is AMR-WB_12.65 with 12.65 kBits of data per second, which is intended for pure voice signals. In the case of more complex signals, the bit-rate-intensive variants are switched up accordingly. If the transmission conditions are poor, you can fall back on the "narrowband" playback variants with 8.85 or 6.6 kbit / s. With a speech pause detection , the transmission can be limited to so-called "Silence Descriptor" (SID) with 1.75 kbit / s ( interrupted transmission , DTX) which only contain parameters for controlling a comfort noise generator in periods without an actual useful signal .

The method works with blocks of 320 samples (= 20 ms). With an additional 5 ms of lookahead, the result is a codec latency of 25 ms. According to the manufacturer, the calculation effort (complexity) is 38.9  WMOPS .

Enhanced Voice Service (EVS)

EVS is being developed for broadband voice connections in third and fourth generation mobile radio systems. Various additional data layers can be added to a downwardly compatible AMR-WB layer. Means replication of the signal from the AMR-WB layer a supplementary high-frequency signal is synthesized. For this purpose, suitable sections of the output signal are selected in a frequency-transformed representation of the signal for different subbands of the supplementary signal to be synthesized and combined at an adapted volume or - alternatively or additionally - the overtones of the basic signal are generated for very tonal signal components by appropriately scaled sinusoids . The encoder determines the parameters for controlling the reconstruction process and transmits them in an extension data layer. This adds another octave to the spectral range to create a super broadband signal (up to 14 kHz). With another similar additional data layer, this can be expanded by a further octave to cover the full human hearing range ("full band"). Further possible additional data streams can contain parameters for the reconstruction of further audio channels .


AMR-WB is used in classic third-generation mobile communications in accordance with the purpose for which it was developed. In the summer of 2006, Ericsson carried out an AMR-WB operational test in the T-Mobile UMTS network in Germany with selected customers in the cities of Cologne and Hamburg. All Ericsson BSCs in the T-Mobile network have been prepared for AMR-WB since the end of 2008, and since November 2011 this has been available to end customers across the board. After Telekom and Vodafone, E-Plus has also been offering the process under the marketing name HD Voice since March 2014 . O2 was the last mobile phone provider to introduce AMR-WB in March 2015.

As part of the CAT-iq , this codec is also intended for cordless home telephones based on the DECT standard for broadband voice transmission (16 kHz sampling rate). It is also used for IP telephony applications. The process is used by various so-called softphones with SIP and other transmission protocols.


The 3GPP has published a reference implementation for AMR-WB in the source code . VoiceAge provides an experimental version for non-commercial purposes in the source code for Windows operating systems. From the OpenCORE project there is an implementation as free software that is distributed under the conditions of version 2 of the Apache license . FFmpeg first made it possible to use the OpenCORE library since March 2010 and developed its own free implementation of a decoder for the format as part of a programming grant from the Google Summer of Code .


The basic ACELP process was developed in 1989 at the Université de Sherbrooke in Canada. The reason for the development of AMR-WB was the search by the European Telecommunications Standards Institute (ETSI) and the 3rd Generation Partnership Project (3GPP) for a new broadband codec with the ability to adjust the bit rate, the Adaptive Multi-Rate Wideband. After a feasibility study in spring 1999, the standardization process was initiated in the middle of the year. The procedure developed jointly by Nokia and VoiceAge was selected for the standard after the qualification phase in spring 2000 and the selection phase that lasted from June to October in December after practical tests. The specification was then finalized and approved in March 2001. The procedure has been a standard recommended by the International Telecommunication Union (ITU) since January 2001 (G.722.2). The Variable-Rate Multimode Wideband (VMR-WB) method used in the CDMA2000 3G standard was developed based on the example of AMR-WB and is compatible with it.

For LTE (mobile communications generation 3.9 / 4), the 3GPP is striving to achieve a substantial improvement in the voice quality for telephone calls. The Nokia Research Center published a draft of the concept on April 18, 2010.

Web links


  1. Speech Coder ( Memento from January 6, 2012 in the Internet Archive )
  2. a b Wideband Speech Coding Standards and Applications (PDF) ( Memento of October 13, 2007 in the Internet Archive )
  3. a b Media coding for the next generation mobile system LTE ( Memento from September 24, 2015 in the Internet Archive )
  4. Björn Brodersen: Telekom's mobile network largely equipped for HD telephony., May 5, 2011, accessed on March 20, 2014 .
  5. Alexander Kuch: HD Voice in the endurance test: Away with the background noise., November 20, 2011, accessed on March 20, 2014 .
  6. Alexander Kuch: Now also in the E-Plus network: HD Voice ensures better voice sound., March 12, 2014, accessed on March 20, 2014 .
  7. Markus Weidner: Introduction at o2: HD Voice in future in all German networks., March 7, 2015, accessed July 7, 2016 .
  9. The VoiceAge AMR-WB Implementation ( Memento from September 26, 2011 in the Internet Archive )
  11. Technologie de compression de la parole ACELP (French).