AMR, or Adaptive Multi Rate, is an audio codec originally defined by the ETSI standards group for carrying voice traffic over the GSM telecommunication system. The specification was later selected by the 3GPP to become a mandatory speech codec for the W-CDMA system.

The goal when specifying the AMR codec was to improve on the existing good quality but bandwidth intensive EFR (Enhanced Full Rate) codec and the less bandwidth greedy but lower quality HR (Half Rate) codec.

AMR achieves its goals by dynamically adapting its bit-rate allocation between speech and channel encoding, thereby optimising speech quality in various radio channel conditions. This results in voice being encoded in various "modes" varying in the number of kbit/s used. The higher the kbit/s, the better quality the result.

Now for the science bit...

  • AMR operates on 20ms frames at 8 kHz sampling frequency
  • 8 speech coding modes in the range 4.75 / 12.2 kbit/s
  • Seamless mode adaptation
  • Also includes DTX (discontinuous transmission) / comfort noise and lost frame compensation