Audio tweaks pre-encoding for high quality MP3 output

Question

I run an internet radio station. We encode our audio into MP3 at 128kbit, joint-stereo 44.1kHz, with LAME. Obviously there are limits to the quality that can be achieved with this bandwidth... that's not my question.

I would like to know if there is any sort of pre-processing we can do with the audio to improve the sound of the resulting compressed audio, reducing the compression artifacts a bit. I've noticed that some internet radio stations sound better than most. Hot 108 Jams comes to mind. I don't know if they're doing anything special, but my ear tells me it is better than many other stations.

Some thoughts I've had:

• Low-pass filter around 17kHz or 18kHz
• Using a compressor and/or expander to reduce the dynamic contrast

I plan to start experimenting soon. Are there any suggestions you have for things to try?

Again, my goal is to build a signal chain in such a way to reduce the artifacts introduced during the MP3 encoding process. I am hoping to create a source signal that lends itself to being encoded into MP3,while sticking to the original signal as close as possible.

Our source material is varied. We play a wide variety of genres, from a wide variety of media. Mostly everything is digital. Much of our material is in MP3 format already. (Sadly, the smaller labels find this to be an appropriate form of digital distribution.) So, many cases we are re-compressing already MP3 compressed audio.

Also, if it matters, we are currently using the line out of a Mackie 1604VLZ Pro into an Alesis 3630 (for compression and limiting), and then into an M-Audio Audiophile 192 for the stream.

I hope to find a balance between tweaking the signal and leaving it alone, where the resulting 128kbit MP3 stream sounds best, and closest to the source material.

Answer 1

Best I've been able to come up with to this one (and I'm interested in a better answer here myself) is a GearSlutz thread on what (if anything) anyone is doing differently to master for streaming audio. The answer from Lupo is relevant and interesting. He suggests:

Level is important. 'Hidden' overshots will be revealed through coding and decoding. ... Using an oversampled peak level meter to keep any /real/ peaks below zero is probably the best approach.

Giving a specific dB figure is hard to do here, as it will vary a lot between different songs and mixing engineer treatments! If working in the blind, without oversampled metering, a ceiling of around -3dB is probably a good starting point.

As for frequency content:

As for frequency content, I think the only important thing is the good old rule - keep it flatish. Tiny little speakers have virtually no deep bass, but most of them have a hefty bass boost to compensate. The 2.1 setup with a ~80hz 'sub' and tiny sattelites are very common today! I think boosting in the upper bass region to make it sound better on laptops may introduce more problems on other systems like the boomy 2.1 setups. As long as I can get any hint of a deep bass at all on my laptop, I'm satisfied! =)

He also suggests the highest bit depth resolution you can use.

So that jives with the low pass filter you were considering, but goes against using any extreme compression on the track to reduce dynamic contrast.

I can't imagine an internet radio station that doesn't do its own compression and eq'ing. So compressing a track knowing it's going to get compressed again before broadcast doesn't seem worthwhile. I'd err on the side of handing them something that sounds good, dynamic -- don't compress it too much.