The Adaptive Multi-Rate (AMR, AMR-NB or GSM-AMR) audio codec is an audio compression format optimized for speech coding. AMR speech codec consists of a multi-rate narrowband speech codec that encodes narrowband (200–3400 Hz) signals at variable bit rates ranging from 4.75 to 12.2 kbit/s with toll quality speech starting at 7.4 kbit/s.
AMR was adopted as the standard speech codec by 3GPP in October 1999 and is now widely used in GSM and UMTS. It uses link adaptation to select from one of eight different bit rates based on link conditions.
AMR is also a file format for storing spoken audio using the AMR codec. Many modern mobile telephone handsets can store short audio recordings in the AMR format, and both free and proprietary programs exist (see Software support) to convert between this and other formats, although AMR is a speech format and is unlikely to give ideal results for other audio. The common filename extension is .amr. There also exists another storage format for AMR that is suitable for applications with more advanced demands on the storage format, like random access or synchronization with video. This format is the 3GPP-specified 3GP container format based on ISO base media file format.
- Sampling frequency 8 kHz/13-bit (160 samples for 20 ms frames), filtered to 200–3400 Hz.
- The AMR codec uses eight source codecs with bit-rates of 12.2, 10.2, 7.95, 7.40, 6.70, 5.90, 5.15 and 4.75 kbit/s.
- Generates frame length of 95, 103, 118, 134, 148, 159, 204, or 244 bits for AMR FR bit rates 4.75, 5.15, 5.90, 6.70, 7.40, 7.95, 10.2, or 12.2 kbit/s, respectively. AMR HR frame lengths are different.
- AMR utilizes discontinuous transmission (DTX), with voice activity detection (VAD) and comfort noise generation (CNG) to reduce bandwidth usage during silence periods
- Algorithmic delay is 20 ms per frame. For bit-rates of 12.2, there is no "algorithm" look-ahead delay. For other rates, look-ahead delay is 5 ms. Note that there is 5 ms "dummy" look-ahead delay, to allow seamless frame-wise mode switching with the rest of rates.
- AMR is a hybrid speech coder, and as such transmits both speech parameters and a waveform signal
- Linear predictive coding (LPC) is used to synthesize the speech from a residual waveform. The LPC parameters are encoded as line spectral pairs (LSP).
- The residual waveform is coded using algebraic code excited linear prediction (ACELP).
- The complexity of the algorithm is rated at 5, using a relative scale where G.711 is 1 and G.729a is 15.
- PSQM testing under ideal conditions yields mean opinion scores of 4.14 for AMR (12.2 kbit/s), compared to 4.45 for G.711 (μ-law)
- PSQM testing under network stress yields mean opinion scores of 3.79 for AMR (12.2 kbit/s), compared to 4.13 for G.711 (μ-law)