Fact-checked by Grok 2 weeks ago

Adaptive Multi-Rate Wideband

The Adaptive Multi-Rate (AMR-WB) is a wideband speech standard developed by the (3GPP) for enhanced voice quality in mobile communications, and it is codified by the Telecommunication Standardization Sector () as Recommendation G.722.2 for wideband coding of speech signals at around 16 kbit/s. It supports adaptive multi-rate operation across nine modes with bit rates of 6.60, 8.85, 12.65, 14.25, 15.85, 18.25, 19.85, 23.05, and 23.85 kbit/s, encoding 16 kHz sampled audio with a bandwidth of 50 Hz to 7 kHz using (ACELP). This design delivers superior speech intelligibility, naturalness, and quality over traditional codecs (limited to 300–3400 Hz), particularly in noisy environments or at lower bit rates. The development of AMR-WB began in the late 1990s as part of 's efforts to improve speech services for systems, with the algorithm selected in December 2000 following a competitive evaluation. Specifications were approved by in March 2001, and the standard was jointly frozen for both and adoption in July 2001, achieving full ratification as G.722.2 in July 2003 after technical refinements. Key features include source-controlled rate adaptation to optimize bandwidth usage based on channel conditions, integrated (VAD) for efficient discontinuous transmission (DTX), and comfort noise generation (CNG) to maintain perceptual continuity during silence periods. These elements enable robust performance in packet-switched networks, with frame sizes of 20 ms and octet-aligned bitstream formats suitable for (RTP) encapsulation. AMR-WB was initially deployed for wideband telephony in networks under Release 5 and later integrated into Long-Term Evolution () for () services, where it provided a transitional high-quality codec before the introduction of the () codec in 2014. Beyond mobile telephony, it supports multimedia streaming in Packet Switched Streaming Service (PSS) and (), as well as VoIP applications in systems like () endpoints. Licensing for AMR-WB was previously managed through the AMR Family Patent Platform administered by VoiceAge Corporation, ensuring while addressing intellectual property rights for commercial implementations; however, the essential patents expired in 2024, eliminating royalty requirements. Despite the shift toward more advanced codecs like EVS, AMR-WB remains relevant in legacy / infrastructures and low-latency embedded systems due to its balance of quality and computational efficiency.

Overview

Definition and Purpose

The Adaptive Multi-Rate Wideband (AMR-WB) is an audio data compression scheme optimized for in , specifically designed as a multi-rate that supports a frequency range of 50 to 7000 Hz, in contrast to codecs limited to 300 to 3400 Hz. This extended bandwidth enables the codec to capture more of the natural speech spectrum, utilizing (ACELP) techniques for efficient encoding. Developed under the framework, AMR-WB operates on 16 kHz sampled input, processing 20 ms frames to produce high-fidelity speech suitable for packet-switched and circuit-switched networks. The primary purpose of AMR-WB is to deliver high-quality speech while adapting to varying conditions in mobile networks through variable ranging from 6.60 to 23.85 kbit/s, allowing dynamic switching every 20 ms to optimize usage and maintain robustness against errors. This source-controlled variable-rate approach balances audio quality with transmission efficiency, supporting nine operational modes for speech and a comfort mode at 1.75 kbit/s during periods to conserve resources and extend battery life in . By enabling such adaptability, AMR-WB addresses the limitations of fixed-rate codecs in error-prone environments like cellular systems. Key benefits of AMR-WB include enhanced naturalness and intelligibility of speech compared to codecs such as AMR-NB, achieved through its broader frequency coverage that preserves low-frequency presence (50-200 Hz) and high-frequency details for better sounds. This results in reduced and improved listening comfort during extended VoIP or cellular conversations. Overall, AMR-WB elevates voice communication quality in bandwidth-constrained scenarios without excessive computational demands. Created by 3GPP as part of Release 5 specifications finalized in 2001, AMR-WB was introduced to fulfill the growing demand for superior voice quality in emerging 3G networks, marking a significant advancement over second-generation narrowband standards.

History and Development

The development of the Adaptive Multi-Rate Wideband (AMR-WB) codec originated in the late 1990s, driven by the European Telecommunications Standards Institute (ETSI) and the 3rd Generation Partnership Project (3GPP), to address the speech quality limitations of narrowband codecs as 3G systems like UMTS emerged. A pre-study phase began in 1999, focusing on wideband coding needs for improved naturalness and intelligibility in mobile communications. The effort was primarily led by and , in collaboration with other members, building on foundational research in (ACELP) methods for efficient speech compression. Following a competitive selection among candidate codecs, AMR-WB was chosen in December 2000 for its balance of quality and adaptability across . Its specifications were then approved in March 2001 as part of Release 5, marking a key milestone for integration into and networks. First commercial deployments followed around 2007, with initial network tests in 2006 enabling wideband voice in operational UMTS environments. Subsequent evolution included updates in 3GPP Release 7, which enhanced error resilience features to better handle packet losses in circuit- and packet-switched modes, alongside integration with the IP Multimedia Subsystem (IMS) for Voice over IP (VoIP) applications. These improvements, finalized in 2007, expanded AMR-WB's utility beyond traditional telephony. Into the 2020s, AMR-WB has seen adoption in 5G New Radio (NR) for enhanced voice services, serving as an interoperable mode within the Enhanced Voice Services (EVS) framework for Voice over NR (VoNR).

Technical Specifications

Bit Rates and Modes

The Adaptive Multi-Rate Wideband (AMR-WB) supports nine operational modes for speech encoding, with ranging from 6.60 kbit/s to 23.85 kbit/s, allowing adaptation to varying conditions while maintaining quality from 50 Hz to 7 kHz. These modes are: 6.60 kbit/s (mode 0), 8.85 kbit/s (mode 1), 12.65 kbit/s (mode 2), 14.25 kbit/s (mode 3), 15.85 kbit/s (mode 4), 18.25 kbit/s (mode 5), 19.85 kbit/s (mode 6), 23.05 kbit/s (mode 7), and 23.85 kbit/s (mode 8). The lowest two modes (6.60 and 8.85 kbit/s) are designed for temporary use in severe conditions, prioritizing over , while modes from 12.65 kbit/s upward deliver high-quality, near-transparent speech suitable for cleaner links such as those in VoIP or . Each 20 ms speech frame is encoded at the selected mode's bit rate, with the bitstream structured to include a header featuring a 4-bit frame type field that indicates the active mode (values 0–8) and a 1-bit quality indicator, totaling 5 bits for mode-related signaling in the header. Mode switching occurs adaptively at frame boundaries, driven by channel quality feedback, to balance bandwidth usage and robustness; for instance, degraded links trigger lower-bit-rate modes with enhanced error protection via class A bits, whereas stable conditions enable higher rates for improved perceptual quality. To optimize bandwidth during , AMR-WB incorporates Discontinuous (DTX) with integrated (VAD), which suppresses transmission and generates comfort noise using a low-rate Silence Insertion Descriptor () frame at approximately 1.75 kbit/s (frame type 9, 40 class A bits). This VAD-driven adaptation reduces average in noisy environments by up to 50% during non-speech periods, trading minor artifacts for significant power and savings in mobile networks.
Mode IndexBit Rate (kbit/s)Primary Use Case
06.60High error resilience for poor channels
18.85High error resilience for poor channels
212.65High-quality speech
314.25High-quality speech
415.85High-quality speech
518.25High-quality speech
619.85High-quality speech
723.05Transparent wideband speech
823.85Transparent wideband speech
9 (SID)~1.75Silence/comfort noise via DTX
The trade-offs across modes emphasize robustness at lower rates, where fewer pulses and bits (e.g., 132 bits at 6.60 kbit/s versus 477 bits at 23.85 kbit/s) allocate more resources to channel coding, achieving lower frame error rates in error-prone or early links, while higher rates enhance naturalness and bandwidth extension at the cost of increased sensitivity to .

Encoding and Decoding Process

The Adaptive Multi-Rate (AMR-WB) codec employs a structure combining (LPC) for modeling the spectral envelope and (ACELP) for quantizing the excitation signal, with pitch prediction to capture periodic components of voiced speech. This approach enables efficient compression of wideband speech sampled at kHz, processing frames of 20 ms duration. The encoding process begins with pre-processing, where the input signal undergoes a with a 50 Hz to remove components, followed by pre-emphasis using the H(z) = 1 - 0.68 z^{-1} to boost higher frequencies and improve coding efficiency. Next, LPC analysis is performed over a 20 ms windowed frame, yielding 16 coefficients that represent the short-term spectral envelope; these coefficients are derived from the method using the for stability and efficiency. The LPC is mathematically modeled as A(z) = 1 - \sum_{k=1}^{p} a_k z^{-k}, where p = [16](/page/16) and a_k are the LPC coefficients. The residual signal is then quantized using ACELP, which involves an adaptive codebook for pitch periodicity and a fixed algebraic codebook with interleaved fixed pulses for the stochastic component; gains for both codebooks are jointly vector quantized to minimize perceptual distortion. Rate adaptation is achieved by varying the codebook sizes and pulse configurations, allowing flexible bitrate control without altering the core structure. Decoding mirrors the encoding process in reverse: the quantized parameters reconstruct the by combining contributions from the adaptive and fixed codebooks, which is then filtered through the LPC synthesis filter to produce the speech signal. Post-processing includes de-emphasis to reverse the encoder's pre-emphasis, along with a post-filter for perceptual enhancement through and sharpening, and gain smoothing to reduce artifacts. For robustness in packet-switched networks, error concealment techniques interpolate lost frame parameters from adjacent frames, muting or repeating as needed to maintain continuity. A distinctive feature of AMR-WB for wideband operation is its split-band analysis, where the is divided into a low band (0-6.4 kHz) processed via the core ACELP-LPC method and a high band (6.4-7 kHz) generated by shaping the low-band with a dedicated and adding for unvoiced components, ensuring natural extension without excessive bitrate overhead.

Standardization and Configurations

3GPP Specifications

The Adaptive Multi-Rate Wideband (AMR-WB) codec is formally specified in Technical Specification (TS) 26.173, which provides the ANSI-C source code for its implementation, ensuring bit-exact compliance across devices and networks. This specification was initially introduced in Release 5 (2001) and became mandatory for wideband voice services in Universal Mobile Telecommunications System (UMTS) circuit-switched domains and (IMS) multimedia telephony if wideband speech is supported, starting from Release 5 onward, enabling wideband speech at up to 16 kHz sampling for improved audio quality in 3G networks. Supporting specifications include TS 26.190 for the core description, TS 26.194 for the voice activity detector, and TS 26.173 for the fixed-point ANSI-C code, all under ongoing change control to maintain . AMR-WB supports two primary payload configurations in networks: octet-aligned mode, which pads frames to byte boundaries for simpler processing, and bandwidth-efficient mode, which minimizes overhead by packing bits directly without padding, as defined in TS 26.201 for frame structures and . These modes allow flexibility in transport efficiency, with octet-aligned preferred for legacy systems and bandwidth-efficient for packet-switched environments to reduce and usage. Additionally, TS 26.114 mandates support for AMR-WB with Adaptive Multi-Rate (AMR-NB) frames within a , facilitating seamless transitions between and modes during or interworking scenarios in IMS-based services. Error handling in AMR-WB is addressed through frame erasure concealment mechanisms outlined in TS 26.191, which specify algorithms to reconstruct lost or corrupted frames using predictive techniques based on prior speech parameters. Bad frame detection relies on (CRC) applied to Class A bits in the codec's auxiliary information, as detailed in TS 26.190, enabling the to identify and mitigate transmission errors without audible artifacts in up to 5% frame loss conditions typical of mobile networks. This CRC-protected structure ensures robust performance in error-prone channels, with concealment activating upon CRC failure to maintain conversational quality. For operational profiles, designates the 12.65 kbit/s mode as the primary baseline rate in TS 26.190, offering a balance of quality and capacity suitable for default wideband voice in and IMS deployments. Optional modes, ranging from 6.6 kbit/s to 23.85 kbit/s, enable adaptive rate control in packet-switched networks via source-controlled variable-rate operation, allowing dynamic adjustment based on channel conditions or network load as per TS 26.114. Enhancements to AMR-WB were introduced in Release 9 () through TS 26.114 updates, laying groundwork for compatibility with emerging by standardizing payload negotiation in IMS. Further advancements in Release 12 (2014) integrated AMR-WB with the (EVS) via an interoperable mode in TS 26.446, providing backward-compatible improvements in and noise suppression across all nine AMR-WB bit rates without altering the bitstream. As of 2025, AMR-WB remains integral to New Radio (NR) voice services under TS 26.114 (version 18.10.1 or later), supporting VoNR and multimedia telephony with mandatory octet-aligned and bandwidth-efficient modes for end-to-end wideband delivery in Standalone () architectures. In VoNR, AMR-WB acts as a fallback to ensure with / networks.

Integration with Other Standards

The Adaptive Multi-Rate Wideband (AMR-WB) integrates with (IETF) standards through defined payload formats for real-time transport, enabling its use in Voice over Internet Protocol (VoIP) applications. Specifically, RFC 3267 from 2002 outlines the (RTP) payload format and file storage for AMR and AMR-WB, including (SDP) parameters for negotiation in SIP-based signaling, which facilitates interoperability in packet-switched networks. This was later updated and obsoleted by RFC 4867 in 2007, which refines the RTP payload structure for AMR and AMR-WB, supporting octet-aligned and bandwidth-efficient modes while maintaining compatibility with SDP for VoIP sessions. Beyond IETF protocols, AMR-WB aligns with standards from other bodies for cross-domain . The ITU-T Recommendation G.722.2, with initial approval in 2002 and full version in 2003, defines the identical algorithm as AMR-WB, permitting seamless integration with wireline systems that require wideband speech coding at rates around 16 kbit/s. Similarly, the EN 300 176-2 standard for (DECT) incorporates AMR-WB support for wideband speech transmission in cordless systems, ensuring compatibility in local environments. In multimedia contexts, AMR-WB combines with video codecs like H.264 under Packet-Switched Streaming (PSS) specifications, allowing synchronized audio-video delivery in streaming services where AMR-WB handles alongside H.264 video encoding. For browser-based communications, AMR-WB has been supported in since its early standardization around 2011, with RFC 7875 providing guidelines for its implementation to enhance interoperability with legacy cellular networks in calls. AMR-WB also features adaptations for short-range and specialized wireless links. AMR-WB can be supported in hands-free devices through or custom implementations in later profiles (e.g., HFP 1.6+), aligning with bandwidth constraints for wideband speech over links. In satellite communications, minor modifications enable its use in systems like Inmarsat's networks, supporting dynamic rate adaptation for reliable voice transmission under variable channel conditions. Additionally, AMR-WB maintains full compatibility with the AMR through mode switching in dual-mode devices, allowing seamless transitions between bandwidth modes without interrupting the call.

Implementation and Deployment

Network and Device Usage

The Adaptive Multi-Rate Wideband (AMR-WB) codec has been widely deployed in third-generation (3G) Universal Mobile Telecommunications System (UMTS) networks since its standardization in 2001 as part of 3GPP Release 5, enabling high-definition (HD) voice services over HSPA infrastructure. By 2016, 130 operators had implemented AMR-WB on 3G/HSPA networks across 88 countries. In fourth-generation (4G) Long-Term Evolution (LTE) networks, AMR-WB integration began with Voice over LTE (VoLTE) launches around 2010, supporting HD voice through IP Multimedia Subsystem (IMS) architecture to enhance call quality while optimizing bandwidth. For fifth-generation (5G) networks, AMR-WB has been incorporated since 2020 via IMS profiles for Voice over New Radio (VoNR), ensuring compatibility in non-standalone and standalone deployments. Major carriers such as Verizon, which launched HD voice services using AMR-WB in September 2014, and Vodafone, with rollouts in Germany in July 2013 and the UK in September 2014, have adopted it to deliver superior audio in mobile calls. AMR-WB is integrated into a broad range of devices, including smartphones running since version 2.3 (, released in 2010) and devices starting with the in 2012, which natively support the for encoding and decoding. In VoIP applications, AMR-WB is utilized in systems like () endpoints to enable HD voice over mobile data connections, adapting bit rates from 6.6 to 23.85 kbit/s for efficient transmission. Embedded systems, such as automotive hands-free kits, leverage AMR-WB over for wideband speech, improving clarity in vehicle environments through standards like the Hands-Free Profile (HFP). A pivotal event in AMR-WB adoption was the world's first commercial HD voice service launch by in in September 2009, marking the initial widespread rollout on networks. By 2016, AMR-WB-enabled HD voice had reached 164 operators globally, supporting over 370 compatible devices and serving tens of millions of users, with reporting 27 million VoLTE subscribers utilizing the codec by late 2015. As of 2023, with global unique mobile subscriptions exceeding 5.4 billion—predominantly on smartphones capable of HD voice—AMR-WB underpins voice services for a significant portion of users, though exact figures vary by region. As of 2025, AMR-WB remains integral for HD Voice in global networks, with the market projected to reach USD 6.04 billion by 2032. Deployment of AMR-WB addresses key challenges in , including through fallback mechanisms to AMR codecs in or circuit-switched networks when wideband support is unavailable. Additionally, its multi-rate design enhances power efficiency in battery-constrained devices by dynamically adjusting bit rates to match network conditions, reducing computational load and energy consumption during calls.

Performance in Real-World Applications

The Adaptive Multi-Rate Wideband (AMR-WB) demonstrates strong performance in real-world applications, particularly in and VoIP systems, where its range enhances speech naturalness and intelligibility. In subjective listening tests, AMR-WB operating at 12.65 kbit/s achieves a (MOS) of approximately 4.95 under clean conditions, significantly outperforming codecs like AMR-NB at 12.2 kbit/s, which score around 3.51 on the MOS scale. This improvement stems from AMR-WB's extended of 50–7000 Hz, providing greater perceived quality even at comparable . Objective evaluations using Perceptual Evaluation of Speech Quality (PESQ) further confirm this, with modes yielding scores exceeding 3.5, often reaching 4.04 at 12.65 kbit/s in controlled tests. In practical scenarios such as calls, AMR-WB excels in environments due to its robust handling and adaptive rate switching, which maintains higher speech quality amid background like or crowds. For instance, evaluations in hands-free car kits show that AMR-WB modes preserve intelligibility across various types, including and impulsive sounds, outperforming alternatives by reducing quality degradation under adverse conditions. In group conferencing applications, AMR-WB's encoding contributes to clearer audio transmission, minimizing perceived distortions in multi-party sessions over networks. Its integration in video conferencing platforms leverages the codec's efficiency to support higher-fidelity voice alongside video, enhancing overall in bandwidth-constrained settings. Despite these advantages, AMR-WB imposes a higher computational load compared to codecs, requiring roughly 2–3 times more processing power on resource-limited devices, which can strain older hardware during real-time encoding and decoding. Additionally, in VoIP deployments, AMR-WB exhibits sensitivity to without effective concealment mechanisms; studies indicate that unmitigated losses above 1–3% can degrade PESQ scores by up to 0.5 points, though built-in error recovery features mitigate this better than fixed-rate alternatives. Comparatively, AMR-WB at low bit rates (e.g., 12.65 kbit/s) surpasses G.711's performance in terms of quality-to- efficiency, delivering clarity at a fraction of G.711's kbit/s fixed rate while enabling 20–30% overall bandwidth savings through adaptive multi-rate operation and silence suppression in mixed network conditions. In environments, AMR-WB contributes to improved call reliability, with evaluations showing reduced sensitivity to transient impairments that lower drop rates by up to 15% relative to modes in high-mobility scenarios.

Licensing and Tools

Licensing Framework

The licensing of Adaptive Multi-Rate Wideband (AMR-WB) technology is managed by VoiceAge Corporation, which administers a comprising over 150 granted patents declared essential to the standard. This pool aggregates intellectual property rights from contributors including , , France Telecom/Orange, and VoiceAge itself, facilitating unified access for implementers. VoiceAge was established as the licensing entity following the codec's development, with the AMR-WB formally launched in 2010 to streamline fair, reasonable, and non-discriminatory (FRAND) licensing. The model for AMR-WB is based on per-unit fees for encoders and decoders implemented in end-user devices, with rates tiered by shipment volume to encourage widespread . In , the overall pool was set at $1 per , with VoiceAge's portion determined by a weighting system that allocates shares proportionally to essentiality. Comparable structures for related codecs, such as EVS, feature volume-based rates ranging from $0.22 to $0.40 per device, reflecting similar per-unit economics for AMR-WB bundles. Open-source implementations of AMR-WB exist, such as those based on reference code, but any use, especially commercial, requires compliance with licensing from the AMR-WB to avoid infringement. Licensing agreements adhere to 3GPP's FRAND commitments, requiring essential patent holders to offer worldwide, non-exclusive licenses on reasonable terms without . Cross-licensing options are commonly available for bundled AMR-NB and AMR-WB implementations, allowing implementers to consolidate agreements for multiple codecs. These terms ensure within ecosystems while protecting intellectual property rights. Key events in the licensing framework include the 2010 patent pool launch, which resolved fragmented negotiations among patent owners, and subsequent litigation such as VoiceAge v. (2012), where courts upheld royalty obligations for AMR-WB use in software products. Another significant case, St. Lawrence Communications v. (2015 onward), addressed FRAND compliance for AMR-WB in mobile devices, reinforcing pool-based royalty structures. Compliance with the framework mandates declaration of essential patents to for , as outlined in ETSI/ intellectual property rights policies. Licensees are subject to royalty reporting and periodic audits to verify volumes and adherence to FRAND terms, ensuring transparent enforcement across global deployments.

Software and Development Tools

The reference of the Adaptive Multi-Rate Wideband (AMR-WB) is provided by through ANSI-C specifications. The fixed-point , optimized for resource-constrained systems, is defined in TS 26.173, utilizing 2's complement signed integers for efficient computation on platforms like mobile devices. The floating-point , suitable for higher-precision applications, is specified in TS 26.204, enabling verification of behavior in environments. Both versions support all nine AMR-WB modes and include functions for , discontinuous transmission, and comfort noise generation to ensure consistent performance. Open-source libraries widely incorporate AMR-WB for audio processing and real-time applications. FFmpeg provides native decoding support for AMR-WB, with encoding available through external libraries like libvo-amrwbenc, facilitating integration in multimedia workflows since the mid-2000s. The Open Source Project includes AMR-WB encoders and decoders in its media framework, supporting bit rates from 6.60 kbit/s to 23.85 kbit/s within and AMR container formats across all Android versions. WebRTC implementations leverage AMR-WB for interoperability in real-time communication, as outlined in IETF guidelines, allowing seamless encoding in browser-based voice calls and legacy network bridging. Development tools for AMR-WB focus on licensed SDKs and conformance validation. VoiceAge, as a key patent pool administrator, offers encoder and decoder software implementations compliant with 3GPP standards, enabling developers to integrate AMR-WB into VoIP, streaming, and systems. For testing, 3GPP provides digital test sequences in TS 26.174, which verify bit-exact compliance by comparing output against reference frames for all codec modes and error conditions. These sequences support automated conformance suites, ensuring implementations handle mode switching and frame errors without deviation.