AMR-WB Speech Audio
AMR-WB is a wideband Adaptive Multi-Rate speech codec using 16 kHz sampling to carry a broader voice spectrum than AMR-NB.
The “wideband” in AMR-WB means that its speech representation retains a broader voice spectrum than AMR-NB. It is also standardized as ITU-T G.722.2, a distinct codec from G.722.
At 16 kHz, each 20 ms frame corresponds to 320 sample positions. That greater time-grid capacity helps preserve upper-frequency speech cues, while the codec still uses a lossy voice-model reconstruction.
Use more coded bits for the same speech duration
| AMR-WB mode endpoint | Bits per 20 ms active-speech frame |
|---|---|
| 6.60 kbit/s | 132 |
| 23.85 kbit/s | 477 |
There are nine speech modes between those endpoints. The algebraic CELP reconstruction uses the selected parameters/excitation to create the frame's speech samples. A higher mode supplies more information within the same time interval; it does not make the speech play more slowly.
The sample rate's 8 kHz Nyquist boundary is a property of the grid. The codec's defined filtering and coding determine its actual preserved band. Grid capacity and recording bandwidth should therefore be read separately.
Account for a stored AWB file
RFC 4867 stores active-speech frame data in whole bytes, with a frame-header octet. At the 23.85 kbit/s mode, 477 bits need 60 bytes when padded. The frame occupies 61 bytes including its header.
At 50 such frames per second:
61 × 50 × 8 = 24,400 bit/s.
That is above the nominal 23,850 bit/s because of padding and framing. The initial file header adds a small further amount; silence modes or 3GP packaging can change the complete file rate.
Compare speech representations meaningfully
An older narrowband recording can sound dull next to a genuine AMR-WB source because frequency cues were never retained in the older source. Merely converting that old recording to an AMR-WB or high-rate PCM file cannot supply them.
Conversely, an AMR-WB source can legitimately reach an Emby player as resampled PCM after decoding. The output label describes its current representation; the actual source's codec and coded band explain what was available to reconstruct.
For an AWB or 3GP recording, keep those observations with the active mode, channel count, and full duration. They distinguish a storage-rate calculation, a decoder-format issue, and the expected character of a compressed voice recording.