When exporting to a compressed format, almost every program offers a menu with acronyms like CBR, VBR and ABR and a series of values in kbps. Picking at random leads to files that are too large or, worse, to lower quality than you could get. Let's see what really changes between the three modes and how to find your way.
What bitrate is
Bitrate is the amount of data used to represent one second of audio, expressed in kilobits per second (kbps). Multiply it by the duration and you get the file size:
size (MB) ≈ kbps × seconds / 8 / 1000
A 45-minute episode (2,700 seconds) at 128 kbps takes about 128 × 2,700 / 8 / 1000 ≈ 43 MB. The same episode at 64 kbps takes about 22.
CBR: constant bitrate
With CBR (Constant Bit Rate) the encoder uses the same amount of data for every fragment of audio, whether it is silence or a dense orchestral passage.
Advantages: - perfectly predictable file size; - maximum compatibility with older players and systems; - simpler duration calculation and seeking in some formats.
Disadvantages: - wasted data on simple passages; - possible artifacts on complex passages that would need more bits.
VBR: variable bitrate
With VBR (Variable Bit Rate) you set a quality level and the encoder allocates bits according to how difficult the material is: few on silence and simple sounds, many on transients and rich sounds.
Advantages: - more consistent quality throughout the track; - for the same perceived quality, files are often smaller than CBR.
Disadvantages: - final size cannot be predicted precisely; - on some very old players, duration or seeking may be inaccurate, especially with MP3 if the file lacks the proper header information.
ABR: average bitrate
ABR (Average Bit Rate) is a compromise: the encoder varies the bitrate as in VBR but tries to stay close to a fixed average. Size is almost predictable and quality is better than CBR at the same value.
Comparison
| Mode | You set | Size | Quality | When to use it |
|---|---|---|---|---|
| CBR | Fixed kbps | Exact | Variable | Fixed-bandwidth streaming, maximum compatibility |
| VBR | Quality level | Unpredictable | Consistent | Personal archives, music |
| ABR | Average kbps | Nearly exact | Good | When you need a space ceiling |
Starting values
The following numbers are reasonable starting points, to be checked by listening:
- Mono speech, podcasts: with MP3 between 64 and 96 kbps mono; with AAC you can go a little lower for the same quality. A spoken podcast does not need stereo, which would double the data with no benefit.
- Speech with background music: 96-128 kbps, possibly stereo if the music matters.
- Music: with AAC or MP3, between 192 and 256 kbps stereo, or a high VBR level.
- Real-time communication with Opus: much lower values are enough for an intelligible, natural voice.
Mono or stereo
For a single voice, mono is almost always the right choice. Exporting a mono signal as stereo means encoding two identical channels; with joint coding the encoder notices and wastes little, but it is still less efficient than a mono file at half the bitrate.
Common mistakes
- Exporting at very high bitrates just in case: beyond a certain threshold quality no longer improves perceptibly and the file just gets heavier.
- Re-encoding at a higher bitrate: converting a 96 kbps MP3 to 320 kbps recovers nothing.
- Overlooking the sample rate: at low bitrates some encoders resample or cut the highs; this is normal, but it is worth knowing.
- Not listening to the result: numbers help, but the final check should be done on headphones on the difficult passages.
In my work, for spoken content, I prefer a mid-to-high quality VBR mode in mono, and I always check the episode on a player other than the computer's, to rule out compatibility problems.
In short
- Bitrate determines how much data each second needs and therefore the file size.
- CBR is predictable and compatible, VBR offers more consistent quality, ABR sits in between.
- For speech, mono at a moderate bitrate is almost always enough.
- Raising the bitrate of an already compressed file does not improve quality.
- Recommended values are starting points: the final check is done by listening.
