128K AAC Plus would be fine. Unless you have great speakers, you wont notice the difference between 128 and 256.. except the owner will see a higher bill for streaming.
Psychoacoustics is a contemptible subject because some people hear its artifacts at certain bitrates/settings while others don't -- even when both people's hearing tests the same. All I can say is that I consistently hear the difference between those bitrates on average speakers, and that if allowed to do instantaneous A/B comparisons, they even stand out to me with some speech and music on absolutely cruddy speakers. Then again, I'm one of those unlucky individuals for whom Voltair's watermarking enhancement has made PPM markets unlistenable.
There are other considerations, too. For one, many people use equalizers or treble controls. This means that when you employ "streaming cost optimized" bitrates that only barely conceal coding artifacts, the changes made by your listeners to your audio's spectral balance can dramatically unmask those artifacts. This can he worsened by any radios or software streaming players that process your audio's dynamics in any way.
Another consideration is decoder quality variability. Many people in the industry have long been keenly aware that encoders vary widely in their quality levels. For example, a 128 kbit/s MP3 made by specific LAME versions with specific encoding parameters can sound surprisingly good, while Xing codec-produced MP3s sound horrible even at their highest bitrates. But often overlooked is that different
decoders -- especially older ones built from older codec library code bases -- produce varying quality output given the same bitrates. So, sending lossy audio whose artifacting is only barely concealed through inferior decoders can also result in the sudden audibility of artifacting ... which if equalized or processed can become even more objectionable.
The general rule of thumb I've followed from personal experience with different musical genres, hardware setups, and software decoders, has been to make MP3 encodings at a minimum of 256-320 kbit/s, AAC-LC encodings at a minimum of 192-256 kbit/s, and Opus encodings at a minimum of 160-192 kbit/s, to achieve "transparency under duress." And there seems to be an industry consensus to these rates as well, with, for example, iTunes selling all its music at 256 kbit/s AAC-LC, and with the premium versions of Spotify and Youtube Music streaming at 256 kbit/s in AAC-LC or 160 kbit/s Opus. Using tools like yt-dlp to peek under the hoods of different streaming platforms, I only see 128 kbit/s AAC-LC in use by the non-premium versions of most services. (FYI: Youtube internally labels 128 kbit/s AAC-LC "medium" quality.) I haven't bought any downloadable music from Amazon in a while, come to think of it, but when I last did, everything they were offering was also at 256 (or maybe 320 kbit/s) in the MP3 format. Even in the warez scene, where all the young ears are, the former norm of 128 kbit/s has been deemed something of a pariah relic of the Napster days. 320 kbit/s is now their universal standard for ripped CDs.
In any case, my logic is simple. If radio is to compete with subscription music streaming services, why use bitrates any lower than the ones your competitors do?
I used Starlink and 128K AAC plus to feed KSKO audio from a music festival at a park.. it was flawless and sounded GREAT
Be careful with that. With HE-AAC (a.k.a. AAC Plus), the boundary between the real audio and the SBR syntho-treble region differs in frequency depending on the encoder you're using, and even on whether you're using version 1 or 2. At bitrates like 128 kbit/s, some encoders will produce real audio up to 16 kHz and SBR above that. Because of the limitations of FM, this backfires, as bits that would have been used to encode the transmittable <16 kHz portion of your audio get wasted being diverted to encoding SBR cues that won't even reach your transmitter. With some other HE-AAC encoders I've tested, on the other hand, such as the aac_he profile of ffmpeg's libfdk_aac codec library (which is used to encode a tremendous amount of internet-based streaming material), the boundary for the real audio versus the SBR syntho-treble never goes higher than 8600 Hz when encoding in v2 mode, no matter how high your bitrate setting. That means the HE-AAC SBR will end up making your audio sound
worse (more gritty and lacking in treble tonality) with some encoders at bitrates above ~96 kbit/s than if you just used straight AAC-LC.
Long story short: make sure to carefully compare
your encoder's AAC-LC audio quality to its HE-AAC audio at that bitrate, if you haven't already. It's kind of instinctive to think HE-AAC will improve things at higher bitrates since it does so at lower bitrates. But in general, 80-96 kbit/s is the consensus I see around the internet for people coding audio with or without SBR, and my own ears have always agreed.
