Need advice on audio levelling for tracks on my stations playlist

I've been running a Shoutcast station on Ubuntu for years. It is a very simple setup - just a long playlist pointing to MP3 files stored on the server which are played one after another.

I'm using sc_trans and sc_serv because they are so stable, I almost never have to SSH in to the server to restart anything; it goes for months without needing attention.

I am thinking of going to liquidsoap for my transcoder, because I have been lead to understand that I can do some audio processing in liquidsoap - I want to level the audio from the tracks, essentially I want to add the DSP equivalent of a compressor / limiter. All of the tracks are spoken word so there's not much in the way of dynamics, but some tracks are very low in level- I suppose I could run them all through some analysis and get replaygain numbers for each track, but there are a few thousand tracks so I think I'd rather have liquidsoap doing level adjusting on the fly instead.

Does anyone have any advice for me? I've played with liquidsoap some, but am by no means an expert. I have someone who is helping me who is more of a linux wonk and I am hopeful we will be able to figure out the liquidsoap scripting language enough to get this job done - but maybe someone out there has already done thins and can share their script? Or, if anyone knows of a good forum for advice on using liquidsoap, I'd like to know where I could go for answers if I get stuck...

THanks

Levelling audio is a bit of a black art. the problem is ofte that a track needs different levels at different points. Eg the intro is quiet and needs boosting, but the middle of the song is ok or maybe even topping out. I've got a compressor / limiter doing some of this in hardware, but it's not 100% effective. The best way I've found is when impotrting tracks from CDs etc, import them all into Audacity first and use the built in compressor to bring up the low parts and limit the peaks. Audacity works really well for this. I know it's a lot of hassle to do that for every track, but your station will sound a lot better for it.

My content is all spoken word, and is already quite compressed- just not normalized. I have over 2,000 tracks... pulling each one into Audacity ( or Adobe Audition, which I like) - that's too much work.

Get an MP3 Normalizer app. Automated batch file levelling in a jiffy.

Yes I am considering that. It's a viable approach.

I'm still interested to see what can be done in DSP on-the-fly, however, just to see what's out there, how it's done.

So if I've understood it right, the audio level of each track is consistent all the way through, just some tracks are louder than others?

If that's the case is there a way of changing the broadcast level of each track in the playout software. We use Rivendell and each track can be adjusted in the library at any time. We do it track by track when adding new contentand have things like beds set deliberately low.

I have found a gain-leveling utility that can churn through folders of MP3 files and normalize each one to a set level, so I will just get all my content files in the right loudness ballpark so that when they play in sequence some are not whisper quiet whilst others are shouting. This process seems to result in a happy medium.

After all the tracks are processed I load them back up onto the servers.

Thanks