Plosive repair does exist. Towne notes that pops can be tackled with selective EQ or a dedicated de-plosive processor, but that’s damage control on a sound you could have avoided by turning your head five degrees. Ten minutes checking where you stand in front of the mic<\/a> pays for itself every session after.<\/p>\n The slowest way to work is to record everything, mistakes and all, then hunt for the good takes afterwards. The fastest way is to fix each fluff at the moment it happens, in the same file, with the same tone and the same distance. That’s punch and roll.<\/p>\n Audacity has it built in, under Transport, Recording, Punch and Roll Record<\/a>, and the manual describes its purpose as correcting errors easily during the course of a recording session. You stop, click back to a point before the mistake, and hit Shift and D. The program deletes what came after that point, plays you the audio leading up to it, and drops into record exactly at the splice.<\/p>\n Two defaults matter. The pre-roll, which is how much audio you hear before the punch, defaults to five seconds, which is about right for matching your own energy. The crossfade at the join defaults to ten milliseconds, which is short enough to be invisible and long enough to hide a click. The manual adds one line that people skip: “It is recommended to use headphones when using Punch and Roll recording.” Without them, the pre-roll leaks back into the mic and you splice your own playback into the take.<\/p>\n Other long-form editors have their own version of the same idea. The principle travels: pick up a sentence early, ride in on your own momentum, and you’ll never hear the join.<\/p>\n \u00a9 The Voice Realm<\/p><\/div>\n Editing goes wrong when everything happens at once. You start cleaning noise before you’ve decided which takes survive, then you level a file you’re about to cut in half. Separate the work into passes and each one gets faster.<\/p>\n One: cut for content.<\/strong> Play it through and keep the read you want. Nothing else. No noise tools, no levels, no tidying. If you punched and rolled while recording, this pass is almost empty already.<\/p>\n Two: fix the joins.<\/strong> Listen only to the splices. A join that sounds fine at full speed will often reveal a clipped breath or a doubled consonant when you loop it twice.<\/p>\n Three: deal with noise, if it needs dealing with.<\/strong> More on this below, including why the answer is usually “less than you think.”<\/p>\n Four: breaths, clicks and the rest.<\/strong> Mouth noise, a chair creak, a page turn, the neighbour’s dog in the gap before a sentence.<\/p>\n Five: level, top and tail, export.<\/strong> Loudness last, because every cut you make before it changes the measurement.<\/p>\n Noise reduction is the tool people reach for first and regret most. The Audacity manual is blunt about the trade: “Use the lowest value that reduces the noise to an acceptable level. Higher values than necessary may make the noise even quieter, but will result in damage to the audio that remains.”<\/p>\n The damage has a sound. The same manual describes the artifacts as “random bursts of very short tones at random frequencies, sometimes called ‘musical noise’, ‘bird song’ or ‘tinkly-bells’.” Once a client has heard that watery shimmer under your voice, they hear nothing else. And there’s no free version of the process: “Reducing noise usually results in some distortion. This is normal and there is nothing you can do about it.”<\/p>\n If you do use it, feed it a proper sample. The manual asks for a region of pure noise, and notes that “A minimum of 2048 samples (0.05 sec. at 44,100 Hz sample rate) is needed,” that “A longer Noise Profile is better,” and that faking it won’t help: “Duplicating a very short sample of noise in order to get a long enough sample for the Noise Profile will not help.” This is the practical reason to record a few seconds of silence at the top of every session, before you say a word.<\/p>\n The better fix is upstream. A hum you can hear on headphones is a fridge, a fan, a hard drive or a light fitting, and turning it off costs nothing. Most of the culprits show up in the same short list of problems that keep appearing in home recordings<\/a>.<\/p>\n Stripping every breath out of a track doesn’t make it sound professional. It makes it sound like a robot with good diction. A person who never breathes is unsettling to listen to for more than a minute, and long-form listeners feel it fastest.<\/p>\n The useful move is to reduce rather than remove. Pull a loud gasp down until it sits under the words, keep the small ones, and only cut a breath when the gap around it is doing nothing. Where the breath sounds like a wet click rather than air, that’s mouth noise, and that’s the thing to fix.<\/p>\n Platforms do police this. ACX requires that files are “free of extra sounds like plosives, microphone pops, mouse clicks, excessive mouth noise, and outtakes,” which is a fair description of what a listener notices even when they can’t name it.<\/p>\n Here’s where home recordings get rejected. Not for the performance, for the meter. ACX publishes the bar every audiobook file has to clear in its audio submission requirements<\/a>, and the same numbers are a sane target for narration work generally:<\/p>\n Broadcast and video work measure loudness differently, in LUFS rather than RMS. The European Broadcasting Union’s loudness pages<\/a> summarise the standard this way: “Basically EBU R 128 recommends to normalize audio at -23 LUFS \u00b10.2 LU (\u00b11 LU for live programmes).” The Audacity manual notes the regional split and why anyone cares: “the level for television in the US is normally -24 LUFS and the EBU (European Broadcasting Union) recommends -23 LUFS,” and adds that “a television network can get its broadcast license revoked for a violation.”<\/p>\n You don’t need to memorise all of that. You need a meter in your editor and the habit of checking the number before export instead of after a client emails you about it.<\/p>\n Silence isn’t silence. Digital zero at the start of a file sounds like the audio dropped out, so platforms ask for a little of your room instead. ACX wants “between 1 and 5 seconds of room tone at the beginning and end,” and caps it: “Keep room tone less than 5 seconds.”<\/p>\n Record it properly. Stand where you stood for the read, stay still, and capture ten seconds of the room. Room tone borrowed from a different session, a different mic position or a different time of day will sit oddly against the take, and that mismatch is audible even when the level matches.<\/p>\n Match the format to the job rather than defaulting to one export for everything. Video and broadcast work generally wants WAV, and you’ll rarely be wrong with 48 kHz 24-bit. Audiobook work through ACX wants MP3 at 44.1 kHz and 192 kbps constant bit rate. Auditions want whatever the brief asks for, at a size that fits an email.<\/p>\n Name files the way the client asked, exactly. If there’s no instruction, script name, then take or chapter, then your name, with no spaces. And keep the session file plus the raw recording for at least a few weeks. Pickups are normal, and re-recording one line in the same room at the same distance is easy while the setup still exists.<\/p>\n Run the same check every time and it stops being a check and starts being a reflex.<\/p>\n Listen to the whole file once, not scrubbed, on headphones. Then listen to the first and last thirty seconds on a laptop speaker, because that’s where a client presses play. Look at the waveform for spikes you didn’t hear. Confirm the loudness and peak numbers. Confirm the top and tail. Open the exported file, not the session, and play it. The number of files that go out with an edit undone because someone exported the wrong version is higher than anyone admits.<\/p>\n If the edit is fine but the read is the thing you’re fighting, that’s a coaching problem rather than a software one, and Voice Master Academy<\/a> runs courses aimed squarely at that stage. Editing can only tidy a performance. It can’t put one there.<\/p>\n The first few sessions where you work in passes will feel slower than your old habits. Then the fluffs start getting fixed inside the recording, the noise pass disappears because you turned the fridge off, and the whole job collapses into a read, a listen and an export.<\/p>\nPunch and roll turns editing into recording<\/h2>\n
![\"Infographic:](\"https:\/\/www.thevoicerealm.com\/blog\/wp-content\/uploads\/2026\/09\/info-editing.png\")
Do the passes in this order<\/h2>\n
Noise reduction is a last resort, not a first move<\/h2>\n
Breaths are part of the read<\/h2>\n
The numbers somebody will measure<\/h2>\n
\n
Room tone at the head and the tail<\/h2>\n
What you send, and what you call it<\/h2>\n
The ten minute check before you hit send<\/h2>\n
It gets faster, and then it gets invisible<\/h2>\n