Alternative Editing Method

From Librivox Wiki
Jump to: navigation, search

THIS IS A WORK IN PROGRESS, I'll add the pictures and missing links soon -- icyjumbo

Streamlining the Editing Process in Audacity

When I record for LibriVox, I know I am going to make mistakes that I will need to edit out. Over time I have evolved a fairly streamlined, and yet effective set of tactics for recording and then editing. This article is about how I do this. Even if you start out using some of these tactics, I fully expect that you'll develop your own ways of dealing with the problems that you face in your own recordings. I would encourage you to add your own findings to this page or to the page that linked here.

Some of my tactics for editing start with tactics in recording. I'll describe those after I've described what I do when I edit.

Locating and cutting errors

I know I am going to make mistakes when I record. A lot of mistakes. If you've ever seen the video of Scott Brick narrating, in which he reads five or six pages absolutely perfectly, then you have a picture of exactly how I am not. So I mark every error I make with a tongue click. That is, as soon as I notice I have made a mistake, I click my tongue, then I look back in the text to find the previous pause. That is usually the start of the sentence, but it may be some other punctuation. It doesn't matter, as long as there is a pause in the reading. Starting again after a pause makes it much easier to edit the recording. I take a breath, hold it for a moment, and then start reading from that point that I identified. Why do I hold my breath for a moment? Because it creates a nice silent patch which may be useful for editing. Try it. You'll see that it makes the editing process slightly easier.

When I come to edit the file, I'll always clean the noise first and then save a FLAC version of the original, unedited file. It's been a life saver more than once.

Next, I always make sure that I zoom so that exactly the same amount of audio is visible in the editing window. On my laptop, I make sure that about 100 pixels per second of audio shows in the editing window. That means that I learn to estimate a correspondence between what I see and what I hear, and it helps me navigate through the recording. It also helps me make consistent gaps between words, phrases, and paragraphs.

Then I go all the way to the end of the file, and start working backwards looking for spikes in the waveform. They are easy to spot, as they are very narrow and reach all the way to top and bottom of the window. If I'm not sure whether the spike represents a click, I'll play the audio, but that's very rare.

No matter how many times I re-record something I got wrong, I know that I am always satisfied with the last version. By working backwards I know that I will first come across the version that satisfied me, then the previous version(s) that didn't. It's sad, but I quite often get the same phrase or word wrong more than once, so I quite often have to edit out multiple attempts. If I were to work forwards through the file, I'd edit out the first mistake, only to find that the second version was a mistake too, and have to edit that out.

<picture of click in the waveform>

In the picture you can see the click, followed by an intake of breath, followed by silence, and finally followed by the re-start of the recording. I put the cursor in the silence between the intake of breath and the re-start.

To correct a mistake, I want to select some audio that includes the mistake. I need to make sure that the cut ends match up nicely so that you can't hear the join.

Usually I put the cursor right before the re-start. Sometimes, however, I might keep more of the silence, for example, when there is an intrusive breath just before the start of the previous version.

If the mistake is only a short way back, it's easy to see where to put the start (see David's video for an example of this situation).

Sometimes, however, I have to go back more than just a few seconds to find a good place to re-record. In that case I have to page back along the audio. I can usually recognize the right place because the shape of the audio looks similar, but I can check by playing the audio there without disturbing my cursor's placement. Click in the time-bar above the audio to temporarily start replay there, without moving the 'real' cursor.

<picture of clicking in time bar>

When I find the right place, I shift-click in the waveform just before the audio with the mistake. I try to click about the same distance before the mistake as I previously clicked before the re-start. I can check what my recording will sound like with the selected audio removed by simply hitting the 'C' key. I don't have to cut the selection first, which allows me to quickly adjust the selection if I don't like my first selection.

If I need to adjust my selection, I can do it either by shift-clicking in the new place, or by dragging the selection's end-point. The pointer changes to a finger pointing out of the selection.

<picture of dragging end-point of the selection, showing the pointing finger>

One reason I might want to adjust the selection is if the silence before the selection isn't actually silence, but has a loud intake of breath. This is my chance to get rid of it. I grow the selection to take in the nasty intake of breath at the left-hand end. At the other end of the selection, I have a nice piece of silence to replace it with, so I can shorten the selection at the other end to reveal the nice silence.

If I was to cut the selection now, it might have a small click. Audacity provides a simple tool called 'Find zero crossings' to adjust the endpoints of a selection so that the waveforms line up nicely after the selection is deleted. The tool's keystroke is 'Z'. I type 'Del' to delete the selection, and straight away I type Ctrl+S to save the project. I have found Audacity quite robust really, but it does occasionally crash. This habit I have developed of saving the project after every little edit has saved me a substantial amount of re-work, so I think it's well worth incorporating into my editing routine.

When I've edited out the mistake, I continue scrolling backwards through the file until I have found the next error. It all sounds quite involved and intricate, but it very quickly becomes a snappy habit, and makes good edits.

When I have edited out all my mistakes, I cut the beginning and the end of the file to the right lengths, amplify it if necessary, and then save it, first to a FLAC version, then to MP3. Amplification is another article, but the short version is that once I remove the clicks, nothing else should clip, so I can just amplify the whole file so that the maximum volume is 0dB. The amplification value suggested by the Amplify... dialog box will do exactly that. Sometimes, however, it's necessary to do a little more work to get a reasonable volume throughout the whole file.

Summary of the error-removal process

  1. Find a clicker, either your tongue, or your fingers, or a jam-jar lid, or whatever suits you.
  2. When you make a mistake, go back in the text and find a suitable gap, such as a full-stop or a comma.
  3. When you've found it, make a click, take a breath, and hold it.
  4. Start reading again to correct the mistake.
  5. Import the audio, and zoom to show about 10 seconds of audio in the window.
  6. When you see a click in the waveform, put the cursor just before the re-start, and go and find the previous bit of the recording that sounds like the re-start.
    click in the time bar to play a section of the track without disturbing the cursor.
  7. When you find the right place, shift click just before the audio to be replaced.
  8. If you want to hear what it sounds like with the selection cut out, type 'C'.
    If you don't like the selection, adjust it by shift-clicking in a different place.
  9. Type 'Z' to adjust the selection to put its endpoints at zero crossings.
  10. Type 'Del' to remove the selected audio.
  11. Type Ctrl+S to save the project.
Personal tools