Satisfying Sounds in Audacity

I've been working rather hard on this audiobook.

Part of the problem is that I apparently make a lot of "mouth sounds." But I am also running into things like a purring cat, strange vibrations, and mispronunciations.

Now, mouth sounds are supposed to be easy to fix. My mouth sounds are not. I get the normal clicks that a "declicker" can catch, sure, but it doesn't get them all. I also get wet smacking sounds and guttural sounds that aren't so easy to fix. 

Yet, I keep learning more about Audacity and either discovering or inventing new tricks to deal with them. Only a few have been able to get through.

A caveat: I am NOT an expert. I watched some of the experts. I am unwilling to spend an hour on less than 5 minutes of audio.

Here are some of the things I do.

Most of the mouth sounds are not actually during words. They're between words, right before or after a word. So if i copy a section of room silence (just record the empty room—you need this at the start and end of files anyway) and then paste it in place of the section between words (no matter HOW small), you'll take care of 90% of the sounds.

There's actually an easier trick I figured out on my own. There's something called a "low-pass filter." All this means is it turns the volume down (or off) on anything above a certain hertz (i.e. higher "notes") and lets the lower ones pass through. If I put the filter on to catch anything about, say, 150 Hz, and I use that only on the parts where I'm not speaking, it neatly eliminates almost every unwanted sound. What makes this easier is that I can then type ctrl-r and repeat it. No deleting for length is needed and if the room sound changes a little, it sounds much more natural.

But sometimes a sound is RIGHT where I'm talking. Chances are, if I do the trick above to either side of the word, my voice drowns out the rest and you can't hear it. And if it's short enough, and I zoom in far enough (and use "z" so it doesn't make artifacts), just deleting that section can usually remove the sound without really affecting my voice. But there are other things to do.

Audacity's "spectral" tools are neat and can be useful, or they can be useless. By right-clicking on the y-axis, I can change the view to "spectrum." This shows where the sounds are in hertz. Then I select the offending part and work up and down the word using "spectral delete" to figure out just WHERE that little bugger is hiding (if it isn't obvious). Once I find it, I undo everything and then use "spectral multi tool" on it. That often hides the sound without disturbing my voice too much. But this only works maybe 30% of the time. For instance, in the case of my purring cat, her purr is at the same hertz as the bulk of my voice, so anything I do to her purr affects my voice as well.

This is where I look for the "noise gate." The noise gate works kinda like the low- or high-pass filter, except instead of blocking a range of notes, it blocks a range of volume. Since her purr is a lot quieter than my voice, I play around with the settings until it blocks out her but doesn't affect me. Voila!

What if i can't find the sound on the spectrum and a filter would kill too much of the sound? Equalizer! Audacity has a variety of them. the graphic equalizer lets you turn up and down different frequency ranges. They say most unwanted sounds are in the 8k+ range, but a lot of mine are close to 3 or 4. Turning down a band or two for a short section usually doesn't affect my voice enough to be noticed by someone just listening to the book, and if it doesn't outright kill the sound, it can turn it down enough that it gets ignored. "Preview" is really good for this. I use it all the time.

If there's a short sound (like a click) that the declicker didn't get, but it's during part of a word where I simply can't mess around with my voice at all (like a "d"), or if I want to do things "the right way," I can use "repair." Right-click on the y-axis and change the view from amp to dB. Then zoom zoom zoom in. Find the offending section. Select just it or part of it (the tool can only fix really small areas) and use repair until it's all gone. When it works, it is freaking awesome. But so far, it is very hit or miss.

If you're having trouble finding just WHERE a sound is, I have a way. When you push play, the end of a selected section is often left off, but the beginning is always precise. So place the cursor before the sound and push play until you hear it. Then pause. Move the cursor further down and try it again. Keep doing that until you hit a section where you don't hear the sound and you know it is just before that part. Granted, there's often a clicky sound when you push play in the middle of the word, so you have to learn to tune that out.

Of course, there's another awesome way to get rid of an errant sound. I was having trouble with a strange vibration that went throughout an "s." It was a broad enough spectrum that if I removed it using the spectrum tools, my voice would sound like I was in a tin can. It was the same volume as my voice. It wasn't small, so repair wouldn't work. I was at a loss! So... I found another place where I used an "s" and copied it, then pasted that over the offending "s." Lo and behold, I couldn't tell I'd done anything and the vibration was gone. I've been able to use that trick for many letters and even entire words, but I have to be careful because the tone and volume of my voice might change between the original and the new sections. I ALWAYS play it back to make sure it sounds smooth.

What if I realize I left off a bit of a word (like a "d" or "s")? Copy paste from another! But that doesn't work for everything. Sometimes, I slur my words together. I rush. Rushing is bad. "Hehad toooze" instead of "He had to ooze" (don't judge me, it's an example) sounds awful. I have a couple options.

The first is to take some (TINY bit) of the room noise and paste it between the words. However, that will sound terrible. So I use "fade out" and "fade in" on the ends of the two words. Often, that is enough to make it sound like a natural flow. 

The second is a bit harder but sounds even better. Find where the two words should separate. Then make a divot between them. Turning down the volume a little and in increments can work but is time consuming and doesn't sound the best. There's a way to do it with one of the fade tools, but I haven't figured it all out yet (I believe it uses "adjustable fade"). But there's a cheat. I add room noise between the two words, then I fade either end like before. But instead of leaving some of the room noise there, I actually select the very tips of the two sections and delete them WITH the room noise. I gotta use "z" and I need to make sure both parts are at the same volume level (zoom in a lot), but what I end up with is a smooth divot between the words. It's enough that you'll easily differentiate the two words while listening to the book.

Hopefully this will help someone who is in a similar situation.