AMR-NB Speech Audio
AMR-NB is the narrowband Adaptive Multi-Rate speech codec used in mobile voice and recorded speech.
| AMR-NB property | Value |
|---|---|
| PCM sample rate | 8 kHz |
| Coded speech-frame duration | 20 ms |
| Reconstructed samples per channel/frame | 160 |
| Speech coding modes | Eight, from 4.75 to 12.2 kbit/s |
Narrowband Adaptive Multi-Rate is a speech codec used in mobile voice and stored voice recordings. Its modes allocate different amounts of coded information to the same 20 ms speech interval.
Count what a mode actually stores
At 4.75 kbit/s, one 20 ms speech frame contains 4,750 × 0.020 = 95 coded bits. At 12.2 kbit/s, it contains 244 coded bits. The decoder reconstructs 160 speech samples in either case.
The difference is the amount of information describing the voice model and excitation, not playback speed. The speech-coding article explains that reconstruction.
RFC 4867's octet-aligned file storage adds a frame-header octet and packs the coded bits into whole bytes:
| Active-speech mode | Coded bits | Padded payload bytes | Bytes with one frame header |
|---|---|---|---|
| 4.75 kbit/s | 95 | 12 | 13 |
| 12.2 kbit/s | 244 | 31 | 32 |
With 50 active frames per second, the second case uses 1,600 bytes/s, or 12.8 kbit/s, before the initial file header. That explains why a stored-file data rate can exceed the nominal speech-mode rate. Silence-related frames and other containers change the overall accounting.
Interpret the narrowband sound
The 8 kHz grid supports the narrowband speech system. Its restricted frequency range and voice-model approximation can give an intelligible recording a telephone-like character.
A 48 kHz output version can reproduce that signal on another grid, but the original coded bandwidth remains the source of its tonal character. AMR-WB is a different wideband codec with different reconstruction, not a setting that expands the old narrowband recording.
AMR frames can be stored directly or inside 3GP media. For an Emby phone recording, identify that wrapper and channel presentation alongside the speech mode. It tells the reader how to retrieve the correctly timed coded frames without treating the nominal bitrate as the exact file-size rate.