Step 1: Pick an editor and set up a lossless project
The tools below differ mainly in how they edit: Audacity and GarageBand show you the waveform and you cut it; Descript shows you a transcript and you cut the words; Auphonic and Riverside’s Magic Audio process a file you give them. Every audio step on this page can be done in Audacity, which its makers describe as free and open source and which runs on Windows, macOS and Linux, so that is the tool the steps refer to.
| Tool | How it edits | Cost as published | Source |
|---|---|---|---|
| Audacity | Waveform editor and recorder: cut, noise reduction, compressor, limiter, loudness normalization, MP3 export. | Free and open source; Windows, macOS and Linux. | audacityteam.org and the Audacity Manual |
| GarageBand | Apple's music studio app for the Mac, with an Audio Editor and voice recording. | Free on the Mac App Store; requires macOS 15.6 or later. | Mac App Store listing; GarageBand User Guide |
| Descript | Edits audio by editing the transcript; filler-word removal and Studio Sound. | Free $0 with 60 media minutes a month (filler words and Studio Sound Limited); Hobbyist $16/month billed annually or $24 monthly for 10 media hours. | Descript pricing page |
| Auphonic | Processes the audio you give it; features listed include Intelligent Leveler, Noise & Reverb Reduction, Loudness Specifications. | Free for 2 hours of processed audio per month. | Auphonic pricing page |
| Riverside Magic Audio | Applied to a recording on Riverside: removes background noise and corrects loudness and volume levels. | Not quoted: the pricing page showed a staging-environment notice when we fetched it. | Riverside Magic Audio page |
Our advice: Before the opening cut, copy every raw file somewhere you will not edit it, and edit a copy. A raw file you have not touched is an undo that survives a crashed project.
Then make sure you are working in a lossless format. Audacity’s manual is blunt about it: if you import an MP3, edit it and export it as an MP3 again, you lose quality twice, so “never do production work in MP3”; work in WAV, AIFF or FLAC and export MP3 only for final use.
Our advice: If you recorded remote guests on a tool that saves a separate track per person, import each track separately and keep them lined up. One track per voice is what makes steps five and six possible without compromising the other speaker. If you only have one mixed track, the steps still work; you just treat everyone at once.
Step 2: Listen through once and mark
The temptation is to start cutting at minute one. Resist it for one pass.
Our advice: Play the whole recording once, on headphones, without editing, and mark what you hear: the sections you will cut, the moments that should open the episode, noise that comes and goes, a voice that drops in level. In Audacity, a label in a label track is enough; on paper, a timestamp and three words. A list of marks turns the edit into a sequence of decisions you have already made, and it stops you polishing a section you were going to delete anyway.
Our advice: Decide the shape on this pass too. If the strongest two minutes are at minute 34, they may be your opening, or at least the clip you cut later. If the opening five minutes are warm-up, they are usually the obvious cut.
Step 3: Cut for content
Our advice: Cut the big things before the small ones: false starts, tangents, the “can you hear me?” at the top, the section the guest asked you to remove. There is no point removing an “um” from a paragraph that is about to go.
- In Audacity, Audacity’s support pages (written for Audacity 4) say to select a section by clicking and dragging in the waveform, then press Delete or Backspace. Hold Ctrl and scroll (Cmd on macOS) to zoom in and place the edges of a selection precisely. The split tool, the scissors or the S key, divides a clip so you can move a section; trims are non-destructive, so a trimmed clip can be pulled back out.
- In Descript, you highlight the text and press Delete (Backspace on Windows); its help page says deleted content leaves playback and exports but can be restored. “Ignore” strikes the text through instead, so it stays visible in the script without playing, which is useful while you are still deciding.
Our advice: Cut on a breath or a pause, not mid-word, and listen across every cut once. A cut that is invisible on the waveform can still be audible as a jump in room noise or tone of voice.
Step 4: Tighten filler words and long pauses
Now the small things: filler words and long silences. Two tools handle this, and both let you choose how much to take out.
- Filler words in Descript. Descript’s help page says it detects filler words such as “um” and “uh” (in English transcripts only), and its Remove filler words tool lets you handle each one: delete it, delete it and leave a gap to keep the timing, ignore it, or remove it from the transcript only. Its Avoid harsh cuts option skips filler words that cannot be removed without clipping into nearby words. On current plans the tool uses AI credits, and Descript’s pricing page lists it as Limited on the Free plan.
- Long silences in Audacity. Truncate Silence, per the manual, shortens any passage that stays below a threshold you set (between -20 dB and -80 dB) for at least a duration you set. It can cut every detected silence to the same length, or compress the excess silence to a percentage of its original length, which keeps longer pauses relatively longer.
Our advice: Do not remove every filler word or every pause. Take out the ones that interrupt a sentence or pile up, and keep the ones that sound like someone thinking; speech with every pause removed sounds clipped and tiring. Run Truncate Silence on a copy first and listen to a few minutes before you apply it to the episode.
Step 5: Reduce background noise
Audacity’s manual is specific about what Noise Reduction is for: constant background sounds such as hum, whistle, whine, buzz and hiss, including fan noise. It says it is not suitable for individual clicks and pops, or for irregular noise such as traffic or an audience, and that a satisfactory result may be impossible when the noise is very loud, variable, or close in level or frequency to the voice. The two-step method, as the manual gives it:
- Get a noise profile. Select a region that contains only the noise (at least 2048 samples, about 0.05 seconds at 44,100 Hz; longer is better), open Effect > Noise Removal and Repair > Noise Reduction, and click Get Noise Profile.
- Reduce. Select the whole region to clean, open the effect again and adjust the sliders, using Preview. The manual says to use the lowest Noise Reduction (dB) value that reduces the noise to an acceptable level (the default is 6), because higher values damage the audio that remains. Switching to Residue lets you hear what will be removed; if you can hear bits of the voice in it, the settings are too high.
Order matters, per the same page: do any notch filtering or click removal before noise reduction, and compression or other effects after it, not before.
If you would rather not set sliders, Descript’s help page describes Studio Sound as an effect that enhances spoken voice by reducing background noise, echo and other distractions, with an adjustable intensity; it uses AI credits on current plans and is listed as Limited on the Free plan. Riverside’s Magic Audio page describes removing background noise and correcting loudness for recordings on Riverside.
Our advice: Record a few seconds of room noise at the top of every session so this step has a clean profile. Then use the gentlest setting that stops the noise being the thing you notice. Over-processed speech, with the “musical noise” or “tinkly bells” the Audacity manual describes as artifacts, is more distracting to a listener than a little steady hiss.
Step 6: Level the voices
Our advice: Two problems sound alike but are fixed differently: one speaker is louder than another, and one speaker swings between quiet and loud. Fix the first by balancing the tracks; fix the second with a compressor.
Audacity’s Compressor, per the manual, reduces the dynamic range of the audio so it can be amplified further without clipping. Threshold is the level above which compression starts; Ratio is how hard loud parts are pulled down once they pass it; Make-up gain raises everything afterwards. The manual’s worked example: with a -10 dB threshold and a 10:1 ratio, a signal arriving at 0 dB, 10 dB over the threshold, comes out at -9 dB. The manual gives that as an example of reading the graph; it does not recommend a setting for speech.
Our advice: On a separate track per voice, set each voice to sound equally loud first, then compress each track gently, and listen for pumping (the background rising in every pause) as the sign you have gone too far. If you only have one mixed track, compress the whole thing and accept a compromise. Auphonic’s pricing page lists an Intelligent Leveler among its features and says it is free for 2 hours of processed audio a month, which is an option if levelling by hand is where your time goes.
Step 7: Add intro, outro, music and chapters
Our advice: Keep the intro short enough that a new listener reaches the content quickly, use the same intro and outro every episode so the show sounds like itself, and only use music you have the rights to use in a podcast. If the show carries ads, leave a clean pause where a read or an inserted spot will go, so it does not land mid sentence. Mix music under speech low enough that every word stays clear on phone speakers.
Chapters are not part of the audio edit, but this is the moment you know where they fall. Apple’s chapters page says you can provide them in the episode description, through your host with the RSS podcast:chapters tag, or in the file’s metadata; if you provide none, Apple Podcasts can generate them automatically for English episodes of 10 minutes or more (not trailers). Its guidelines for your own chapters:
- In the episode description, start the opening chapter at 00:00:00 and include at least three.
- Provide chapters for episodes longer than 10 minutes, no more than six per hour, and none shorter than two minutes.
- Keep chapter titles to 45 characters or fewer and under five words, in title case, without repeating words across chapters.
Step 8: Set loudness, export and listen once more
The last job is making the episode as loud as the other shows a listener plays before and after it, without distortion. Apple publishes a target. Its audio requirements page recommends that the overall loudness stays around -16 dB LKFS, with a +/- 1 dB tolerance, and that the true peak does not exceed -1 dB FS, both measured per ITU-R BS.1770-5. It says this preconditioning has to happen before encoding, because audio compression algorithms typically don’t modify the loudness and might clip the signal if the true-peak value is not respected. Apple writes LKFS and Audacity writes LUFS; our advice is to read them as the same scale when you set the target.
- Loudness. Audacity’s Loudness Normalization (Effect > Volume and Compression > Loudness Normalization in the manual) sets a LUFS target. Its default is -23 LUFS, the EBU broadcast figure, so change it to -16 for a podcast. The manual recommends it as the final editing step before export. If you edit in mono, note that its “Treat mono as dual-mono” box is on by default; the manual says to switch it off if you need to match meters that measure mono 3 LU lower.
- Peaks. Audacity’s Limiter stops peaks exceeding a threshold, and the manual says its Master Limiter factory preset limits the signal level to -1 dB.
- Export. In Audacity, File > Export Audio and choose MP3; the manual lists mono or stereo, and a sample rate of 32000, 44100 (the default) or 48000 Hz. For RSS feeds, Apple lists 64–128 kbps for mono at 44.1 or 48 kHz, and 128–256 kbps for stereo. Apple accepts MP3 or AAC in a feed and strongly recommends AAC in the MP4 container.
Our advice: Normalize, then limit, then measure the result again; if the limiter has pulled the episode outside Apple’s +/- 1 dB window, repeat with a slightly lower limiter threshold. Leave a little margin under -1 dB, because a sample-peak limiter is not a true-peak meter. Export a spoken-word show in mono unless it has stereo content worth keeping.
A note on Spotify: its loudness normalization page is written for music. It says tracks are adjusted to -14 dB LUFS during playback and that Spotify measures loudness on upload without processing or changing the audio. It does not state a podcast target, so this page does not treat -14 as one.
Our advice: Before you upload, play the exported file from start to end on a phone with earbuds, the way a commuting listener would hear it. Check the opening seconds, every cut you made, the music level under speech and the last few seconds. Then fill in the file’s title and show tags. Mistakes heard here cost a re-export; mistakes heard after publishing cost a re-upload.
From here the episode goes to your host, which adds it to the RSS feed every app reads; the how to start a podcast guide covers hosting, submitting to Apple Podcasts, Spotify and YouTube, and measuring downloads independently from the first episode. PodLink’s free plan carries those download analytics with no card.
When to hand the edit to someone else
Editing is free in cash and expensive in hours. The podcast editing cost guide quotes what the market charges, from AI tools to agencies, with each source and year, and includes the break-even sum for your own time; this guide does not repeat it. Its section on briefing an editor turns the marks from step two into instructions someone else can follow.
PodLink’s editing and production prices are public and the same here as on the service page:
- Audio edit: $250 to set up, waived on a 3-month or longer commitment, then $250 an episode or $900 a month for four. Up to 60 minutes; editing, noise reduction, mixing, music, intro, outro, ads and ID3 tags; one revision.
- Video podcast edit: $500 to set up, then $495 an episode or $1,800 a month for four, with the audio version of the same episode included.
- Turnaround and files: three business days from receiving the file to the edited episode, with rush available; masters, project files and assets stay yours in a shared drive you keep access to.