Achieving uniform level in lengthy spoken word recordings

Question

I often make lengthy recordings of spoken word for Internet distribution - lectures, public debates, etc. The resulting wave files have a highly uneven level - from whisper-quiet (questions from the audience) to moderate (main speakers, close to the mic) to very high (applause, etc.) It is quite tedious to go over a two-hour long wave file, select it bits but bits and manually amplify or de-amplify parts as needed to achieve some uniformity.

I thought this would be an obvious feature to include in various sound editing apps (I'm on Windows), but in fact the opposite is the case. The various "normalize" functions only work across multiple tracks, but not within a single waveform. I use Sony Sound Forge primarily, and Sony's support tells me nothing like I need is available there.

I don't really know much about sound processing, so perhaps there is something obvious that escapes me. Basically, I'm imagining a feature that would "even out" the volume within a single waveform. (Normalization doesn't.) Is there a way to automate this process?

Answer 1

There is an excellent program out there called "The Levelator". All you need to do is drop the raw audio file (WAV or AIFF) onto the app and wait for it to do its job.

From the site:

So what is The Levelator®? It's software that runs on Windows, OS X (universal binary), or Linux (Ubuntu) that adjusts the audio levels within your podcast or other audio file for variations from one speaker to the next, for example. It's not a compressor, normalizer or limiter although it contains all three. It's much more than those tools, and it's much simpler to use. The UI is dirt-simple: Drag-and-drop any WAV or AIFF file onto The Leveler's application window, and a few moments later you'll find a new version which just sounds better.

I use it to level out church sermons (same ... variable volume ... issue you mentioned) and it works wonderfully. Leo LaPorte has mentioned that he uses it for the various TWiT Network podcasts - that's how I first heard about it.

It's free, so why not download it and at least give it a try?

Answer 2

For questions from the audience, there is only one good solution: A microphone that get passed around. If this isn't available I try to remember to always repeat the question for the benefit of the recording when I do my talks.

Other than that, adding compression will help a lot. If you have enough compression the audience questions will be audible too, but you'll get a lot of hiss and noise. In that case do not apply the same compression on the whole file, because you'll get hiss every time there is a pause in the speech, and that's annoying. But first applying one compression level on the whole track to get the main talk correct, and then applying another on the audience question bits isn't usually that much work.

Sony Soundforge 10 sure has an effect for this, it's such a simple effect that I can't imagine they don't include it, and the technical spec page indeed mentions "Dynamics". If you have an earlier version it may be different, but if there is any effects, surely a compressor is included.

Answer 3

I suspect a bit of dynamic range compression might be exactly what you need. Using compressors can be a bit tricky to get dialed in exactly right, but their purpose is to make louder parts closer in volume to the quieter parts, which seems to be what you're trying to do. If the resulting sound is too quiet overall, you can always turn it up after the compression. Some compressors have a "makeup" function that does exactly this.

Perhaps Sound Forge has a compressor function or plugin? As with anything else, check through your recording to make sure you like the results of it. Like with most things, there isn't really a magic bullet.

Answer 4

For a couple of months I worked with a colleague doing audiobook releases. Long winded sessions of volume automation and a slight compression where our best tools when the editing was finished. Now Waves as vocal rider, which can be adjusted for sensitivity or material.

Yes, dynamic compression would be a good way to start (automated process) but having something squashing against it is probably something you don't want to hear.

Most softwares will allow you to automate level, if you are up for the task.

Sound Forge supports VSTs. I haven't used Sound forge for a couple of versions now so I'm not sure if they included anything similar, but at least you can use external plugins of the sort.

Answer 5

Try to see if the 'Leveller' function in Audacity (it's free, so it won't hurt to try) does what you want.. I have been using it for podcasts and it has worked pretty well. But I haven't tested it in very long voice recording. If it does work, then we can look at finding something similar for SoundForge.