Every listener costs you bandwidth for as long as they listen. Bitrate multiplies by concurrent listeners by hours, so the number you pick has a monthly invoice attached.
The defaults
Encoder profiles default to 192 kbps at 48 kHz. That is comfortably transparent for music on MP3 and a reasonable starting point.
The sample rate matches the engine's internal mix format, which avoids an extra resample in the encode path.
Working out the cost
A rough calculation worth doing before you commit: 192 kbps is roughly 86 MB per listener per hour. A hundred concurrent listeners for eight hours is about 69 GB.
Halve the bitrate and you halve the bill. Whether that is worth it depends entirely on what your audience is listening on.
Which codec
MP3 via libmp3lame produces real MP3 frames, served as audio/mpeg. It is self-framing and plays absolutely everywhere. This is the safe default and the right answer for most stations.
AAC uses the Media Foundation encoder with hand-built ADTS framing, served as audio/aac. It gives noticeably better quality per bit — AAC at 96 kbps is roughly comparable to MP3 at 128 — so it is the better choice if bandwidth is the constraint.
WAV/PCM is raw linear PCM, for servers or test rigs that accept it. Not something to stream publicly.
Ogg Opus appears in the codec list but is not implemented, and falls back rather than silently producing nothing.
Practical guidance
- Speech-heavy output tolerates lower bitrates well. 96 kbps AAC is fine for talk.
- Music-led stations should stay at 128 kbps or above.
- If most listening is on phones over mobile data, favour AAC at a lower bitrate over MP3 at a higher one.
- If compatibility with old hardware players matters, MP3 removes an entire class of problem.
Test before you commit
Set up a profile, stream to a test mount, and listen on the devices your audience actually uses. Codec comparisons on studio monitors tell you very little about what a phone speaker on a bus sounds like.