Module-4.4-editing

Session 4

Rescue, Music, and Release — AI Rescue Tools, Music, Export & Loudness

Session 4 · Rescue, Music, and Release — Finishing the Episode

What You Will Learn

  • What AI restoration tools genuinely fix, what they cannot, and how to recognise their artefacts by ear
  • What music is for in a podcast, what makes it sound cheap, and how to place it with fades and ducking
  • Where to find music you are allowed to use — and why "it is only a short fragment" is never a licence
  • How to generate music with AI, and the licensing and transparency questions to ask before publishing it
  • Which file to keep, which file to publish, and which loudness number to aim for
  • What a free path looks like after a paid editor’s trial ends
  • How to finish an episode with a 7-step checklist and test it the way listeners will hear it

📋 Session info

⏱ Duration: about 2 hours 45 minutes, including a 35-minute final exercise and a showcase

👥 Group: 10 participants

🌐 Language: English

🔗 Builds on: Sessions 1–3 — file hygiene, cleaning speech and the filter chain

🔧 Hands-on: everyone finishes, exports and measures a fragment of their own episode

Good recording first. Honest editing second. AI last.

This is the session that takes an episode across the finish line. It meets the AI tools that can rescue a difficult recording — and their limits — then adds music, sets the right loudness, exports the file and checks it on a phone, exactly the way the audience will hear it.

AI Rescue Tools: Power and Limits

AI restoration tools can make a recording with echo or background noise sound remarkably clean in a few seconds. That makes them tempting to rely on. The rule from the recording module still holds: AI processing is the last line of defence, not the first. The room, the microphone position, healthy gain and good technique come first. This section is about meeting that last line of defence in person — and understanding honestly what it can and cannot do.

AI is the last line of defence

Not the first. Room, microphone, gain and technique come first.

A good tool to demonstrate this is Adobe Podcast Enhance Speech, which runs in the browser: you upload a file, it processes the voice, and you can compare before and after. Try it on two kinds of problem. On a recording made in an echoey room, the result is usually startlingly good. On a recording made outdoors, in wind and traffic, the rescue is only partial: the noise disappears, but the voice starts to sound processed.

✅ What AI rescue handles well

  • Steady background noise — a fan, a fridge, a distant hum
  • Room echo and reverb

❌ What it cannot fix

  • Clipping from a too-hot input — the information is simply gone
  • A voice buried under music or another speaker
  • A speaker sitting too far from the microphone

AI improves clarity. It does not recreate detail that was never captured.

Listen for the artefacts

When AI works hard on a difficult recording, it leaves traces. Learning to name them helps you decide whether the "rescued" version is really better than the original.

💧 Watery voice

An underwater, processed quality where the noise used to be.

🔇 Swallowed sounds

Consonants that soften or vanish, especially on the harder rescues.

🎤 Studio-booth tone

An artificial room sound that no longer matches how the space actually sounded.

Free plans of online tools come with limits on file length and on how much audio you can process per day, and these limits change from time to time — check them on your own account before you plan a session around one tool. Many recorders also have AI noise reduction built in, working on the input side. It obeys the same law: good microphone placement and healthy gain beat every rescue tool ever made.

💡 Tip: Always keep the original file. Compare the rescued version with it on headphones before you replace anything — sometimes a little honest noise sounds better than a processed voice.

Other tools that clean audio

Beyond Adobe Enhance Speech, there are three kinds of tool worth knowing by name. Their prices and free tiers change often, so it is more useful to remember what each one does than what it costs.

The professional repair suite

iZotope RX

A deep, paid toolkit with dozens of surgical modules: de-noise, de-reverb, de-click, mouth-click removal, de-plosive and dialogue isolation. What film and radio post-production reach for. Overkill for a beginner, but worth knowing.

Automatic cleanup services

Cleanvoice & Descript

Cleanvoice finds and removes filler words, mouth sounds, stutters and dead air, and shows every proposed cut before export; it works in many languages. Descript lets you edit audio by editing its transcript. Both trade fine control for speed.

Live noise removal

Krisp

Removes background noise and echo in real time during calls and recordings — cleaning at the source rather than afterwards. Handy for remote interviews.

One honest thread runs through all of them: they are conveniences on top of the craft, not replacements for it.

🎓 Trainer Tip: Test your demonstration files through the AI tool before the workshop, so you know in advance that the "impressive" example really impresses and the "struggling" one really struggles. And if a participant’s own recording has a noise or echo problem, ask their permission and use it as a bonus example. Nothing lands like a rescue performed on the group’s own material.

🎙️ Audio note — to be recorded and embedded

Topic: AI rescue — before and after, heard rather than described

Suggested script outline:

  • A short clip recorded in an echoey room, first as recorded, then after AI enhancement — the impressive rescue
  • A short clip recorded outdoors, before and after — the partial rescue, with the artefacts audible
  • A spoken guide to the three artefacts: the watery voice, swallowed consonants and the studio-booth tone
  • One closing reminder: AI improves clarity, but it cannot bring back what was never recorded

[Replace this block with a WordPress audio block after recording]

✏️ Exercise — click or tap to expand ▼

▶ ✏️ Try This With Your Group

Rescued or Ruined?

Training the ear to judge AI processing honestly. Two short recordings — one made in an echoey room, one made outdoors — a browser-based enhancement tool, a laptop and a speaker.

Step 1 — predict (3 minutes): play both originals. Everyone guesses which one the AI will rescue better, and why.

Step 2 — process and listen (10 minutes): run both files through the tool and play each before and after. Everyone notes which of the three artefacts they can hear: watery voice, swallowed sounds or studio-booth tone.

Step 3 — decide (5 minutes): for each clip, the group votes: keep the original, use the rescued version, or re-record. Discuss what would have prevented the problem in the first place.

⏱ Time: about 20 minutes

📋 Alternative — Take-home task: Ask everyone to record 30 seconds of speech in the noisiest corner of their home, run it through a free enhancement tool, and bring both versions to the next meeting with one sentence: better, worse, or just different?

Music in a Podcast

Music gives a podcast identity and structure, but it is easy to overdo. Before choosing a single track, it helps to know what job the music is supposed to do.

Job 1

Intro / outro

Identity: listeners recognise the show.

Job 2

Transitions

Structure: a short sting between parts.

Job 3

Music bed

Mood: quiet music under speech.

Most podcasts made by and with older people need only the first job, and perhaps the second. A music bed under speech is the hardest to get right and the easiest to get wrong.

❌ What sounds cheap

  • One 8-second loop repeating through a whole episode
  • Music louder than the voice
  • A music bed under every single minute

✅ What sounds considered

  • A short intro that fades away under the first words
  • Music used sparingly, where it marks a change
  • Silence — it is a production choice too

Three techniques for placing music

📈 Fade in / fade out

Music rises gently at the start and dies away at the end, instead of starting or stopping abruptly.

🔀 Crossfade

One piece fades out while the next fades in, so the change feels smooth.

🔽 Ducking

The music is lowered while the voice speaks, and comes back up in the pauses.

In most editors, ducking is done by hand: you add fades to the music clip and pull its volume line down where the voice begins. A good intro rises, plays for a few seconds, and then dips under the first sentence so that the listener hears every word.

Legal sources of music

Free music is easy to find; music you are actually allowed to publish takes a little more care. Each of these sources is a good starting point, and each comes with its own caveat — check the licence of every track before you publish.

🎵 BandLab library

Built into the free BandLab editor. Royalty-free within the platform; confirm the terms before using it outside BandLab.

🎵 Free Music Archive

A large catalogue with mixed licences. Check each track’s specific licence.

🎵 Pixabay Music

Free, with no attribution required. Still verify each track before publishing.

🎵 YouTube Audio Library

Free music and sound effects. Read each track’s attribution rule.

The anti-rule

Commercial music is never OK "because it is only a short fragment in the background". Length does not create a licence.

🎓 Trainer Tip: Expect the question "Can I use my favourite song from the radio, just for ten seconds?" Answer it kindly but firmly: no, not without a licence. Then turn the disappointment into a task — many learners enjoy the search for a track that fits their show, and choosing an intro often becomes the moment the podcast starts to feel like their own.

✏️ Exercise — click or tap to expand ▼

▶ ✏️ Try This With Your Group

Cheap or Considered?

Learning to hear what music does to speech. The opening minutes of two or three podcasts with different uses of music, a laptop and a speaker.

Step 1 — listen (6 minutes): play about two minutes of each podcast. Everyone notes which of the three jobs the music is doing: intro, transition or bed.

Step 2 — judge (5 minutes): for each example, ask: can you hear every word? Does the music help or distract? Does anything sound cheap?

Step 3 — decide for your show (4 minutes): everyone writes one sentence: which job music will do in their own podcast — or whether they will use none at all.

⏱ Time: about 15 minutes

📋 Alternative — Take-home task: Ask everyone to browse one of the legal music libraries and choose three candidate intros for their podcast, noting the licence of each. Bring the shortlist to the next meeting and choose together.

AI-Generated Music and the Safe Use of AI

AI tools can now generate a short instrumental track from a written description, such as "warm acoustic intro for a podcast about older people’s life stories, 30 seconds, no vocals". The technology is impressive and easy to use. The lesson that matters more than the technology is about what happens next: whether you are allowed to publish what you generated.

✨ Gemini (Lyria 3 model)

Generates 30-second tracks from a text prompt — roughly the length of an intro. Available to signed-in users aged 18 and over. Every track carries Google’s SynthID watermark, an inaudible marker identifying it as AI-generated.

Before publishing: check Google’s current terms of service for using generated tracks in a podcast.

🎹 ElevenLabs Music

A different engine with a different character — worth comparing with the first. Its music terms set different rights for each plan. On the free plan, attribution is required ("Created in collaboration with ElevenLabs"), use is limited to individuals, and some uses — including streaming and radio — are excluded.

Before publishing: read the plan-specific terms on your own account; for a published podcast, a paid plan may be needed.

Generating is easy

Publishing is a licensing decision. Before any generated track goes into a published episode, read the terms of the tool that made it.

Transparency is good practice too: tell your listeners when the music is AI-generated. Watermarks such as SynthID point in the same direction the whole industry is moving — towards openness about synthetic content.

This is the core of the safe use of AI, and it applies far beyond music. Every AI tool you meet after this course — for voices, transcripts, images or anything else — deserves the same two questions.

💡 Tip: Save a screenshot of the tool’s terms on the day you generate a track. Terms change, and it helps to know which rules applied when you made it.

📜 Question 1: What are the terms?

May I publish this? Must I credit the tool? Does my plan allow it?

👀 Question 2: Are we being open about it?

Do our listeners know which parts were made with AI?

New tool, same two questions. That reflex outlives any single app.

🎓 Trainer Tip: Check the terms of every AI tool yourself in the week before the workshop — free plans and licences change often, and published reports sometimes contradict each other. Then say the key sentence slowly and let it settle: "Generating is easy. Publishing is a licensing decision." Learners remember the demonstration because it is fun; make sure they also remember the rule, because that is what will protect them later.

✏️ Exercise — click or tap to expand ▼

▶ ✏️ Try This With Your Group

Find Your Intro

Every participant sources an intro for their own episode, legally. Laptops, an editor with each participant’s episode open, and either a personal account for an AI music tool or access to a legal music library.

Step 1 — choose a route (2 minutes): generate an intro with an AI tool, or pick a track from one of the legal libraries.

Step 2 — find or generate (5 minutes): if generating, write a prompt that names the mood, the length and "no vocals". If choosing, listen to at least three tracks before deciding.

Step 3 — the two questions (3 minutes): write down the licence or terms of the chosen track, and how you will tell listeners if it is AI-generated. Fast finishers place the intro on the timeline, add a fade and pull the volume line down under the first words.

⏱ Time: about 10 minutes

📋 Alternative — Take-home task: Ask everyone to generate two intros with the same prompt in two different AI tools, compare how they sound, and look up the terms of both. Bring the results and the two answers to the next meeting.

Export and Loudness Targets

An edited episode is not finished until it has been exported in the right format, at the right loudness. Two files and one number are all you need to remember.

💾 WAV — to keep

The lossless original of your finished episode. Use it for further work and for your archive.

🎧 MP3 — to publish

The file listeners get. For speech, common practice is roughly 96–128 kbps for mono and 128–192 kbps for stereo.

Mono or stereo? For a spoken interview, mono is usually enough: the file is smaller, and it plays safely on smart speakers. Stereo earns its place when music matters.

Loudness: the number the platforms ask for

Loudness is measured in LUFS (Loudness Units relative to Full Scale): how loud a file feels on average over its whole length. The numbers are negative — the closer to zero, the louder. A second number, the true peak, tells you how close the loudest moments come to distortion.

-20-19-18-17-16-15-14-13-12
quieterlouder

-19 LUFS

Mono files

A widely used recommendation for mono: about 3 dB below the stereo target.

-16 LUFS

Apple Podcasts

Apple asks for about -16 (±1 dB), with the true peak no higher than -1 dB.

-14 LUFS

Spotify

Spotify normalises playback to about -14; louder files are simply turned down.

The practical rule

Export at about -16 LUFS with the true peak below -1 dB, and your episode behaves well everywhere.

The workflow: export, verify, fix

1 · Export at the target

In the editor

Some editors, such as Hindenburg, let you choose a loudness target in the export settings and deliver the file at that level.

2 · Verify

With a meter

Trust, but check: drop the exported file into a free online meter such as the Youlean Loudness Meter.

3 · Fix if needed

With Auphonic

For audio made elsewhere, Auphonic normalises a file to a chosen target and gives a measurement report. Its free plan covers 2 hours of audio a month.

Then measure again. A loudness number is a useful check, not a substitute for listening: trust the number and your ears together. If the meter says the file is on target but the voice still sounds thin or harsh, the problem lies earlier — in the recording or the filter chain — not in the export.

💡 Tip: Save the exported files inside your episode’s project folder, next to the raw recordings. An export saved to the desktop is an export that gets lost.

🎓 Trainer Tip: Numbers can intimidate. Do not start with the definition of LUFS — start with the experience: play two episodes one after the other, one quiet and one loud, and ask who reached for the volume button. Then explain that LUFS is simply the number that prevents that moment. Most learners need to remember only one figure, -16, and one habit: measure after every export.

✏️ Exercise — click or tap to expand ▼

▶ ✏️ Try This With Your Group

Measure Three Files

Making the loudness number concrete. Three audio files — for example a raw recording, a downloaded episode of a well-known podcast and a file exported by the group — and a free online loudness meter.

Step 1 — guess (3 minutes): play a few seconds of each file. Everyone ranks them from quietest to loudest.

Step 2 — measure (7 minutes): drop each file into the meter and write down its integrated loudness and true peak.

Step 3 — compare (5 minutes): how close is each file to -16 LUFS? Did the ears and the meter agree? Which file would Spotify turn down, and which would sound too quiet next to other podcasts?

⏱ Time: about 15 minutes

📋 Alternative — Take-home task: Ask everyone to measure the loudness of their own latest export and of one episode of their favourite podcast, and bring both numbers to the next meeting.

After the Trial: The Free Path

This course was taught in Hindenburg PRO, a paid editor designed for spoken word, used on a free 30-day trial. When the trial ends, there are two good options: continue with Hindenburg on a paid subscription (check current pricing on its website), or switch to a free tool. The free option shown here is BandLab, which runs in the browser.

✅ What you get, free

  • Runs in the browser on any laptop — nothing to install
  • Non-destructive: trimmed audio comes back by dragging the edge of the clip
  • Multitrack editing, splits, fades and track effects
  • A built-in royalty-free music library

⚠️ The honest trade-offs

  • A music-oriented interface — loops, MIDI and instruments you can ignore
  • No automatic levelling and no loudness-target export (use Auphonic for that)
  • Projects live in BandLab’s cloud, not on your own disk

A quick tour covers everything a podcaster needs: import a file, split it, add fades and track effects, browse the music library, and export. Sessions saved in Hindenburg during the trial reopen if a licence is bought later, so nothing made during the course is lost.

The craft is yours

The cuts, the chain, the ethics, the targets — they all transfer. The editor is just the tool you do it in.

🎓 Trainer Tip: Be honest about money from the start: say which tool is paid, how long the trial lasts and what the free alternative is. Older learners and small organisations often worry that a new skill will lock them into a subscription. Showing a free path before the course ends removes that worry — and prepare a logged-in BandLab account with a small project in advance, so the demonstration takes minutes, not a sign-up procedure.

✏️ Exercise — click or tap to expand ▼

▶ ✏️ Try This With Your Group

Same Edit, Different Editor

Proving that the skills transfer. A short audio clip, a laptop with a browser and a free BandLab account.

Step 1 — import (3 minutes): open the free editor and import the clip.

Step 2 — edit (8 minutes): make the same three moves practised earlier in the course: cut a false start, add a fade-in and a fade-out, and drag a trimmed edge back to show that nothing was lost.

Step 3 — export and measure (4 minutes): export the clip and check its loudness with an online meter.

⏱ Time: about 15 minutes

📋 Alternative — Take-home task: Ask everyone to create a free account in a browser-based editor at home and repeat one edit from the course there. At the next meeting, compare: what felt the same, and what was harder?

The Full Run: Finishing the Episode

Everything comes together in one long final exercise: each participant finishes their episode, adds an intro, exports an MP3, measures its loudness and listens to it on a phone. The checklist below is the backbone of the exercise — and of every episode after it.

1

Speech cleaned

Fillers and false starts out, pauses trimmed, rhythm intact (Session 2).

2

Chain applied

High-pass, noise reduction, compression, EQ (Session 3).

3

Levels balanced

Speakers within about 2–3 dB of each other.

4

Intro placed

With a fade, and ducking under the first words.

5

Export MP3

Keep the WAV; the MP3 is for publication.

6

At the target

About -16 LUFS, true peak below -1 dB, verified with a meter.

7

The phone test

Listen on a phone and on cheap earbuds.

The minimum bar: 3 finished minutes with an intro, exported and measured.

Ambitious participants finish more. Everyone crosses the finish line.

Common last-mile problems

🎵 The intro drowns the first sentence

The ducking is too shallow or starts too late. Pull the music down further, and earlier.

📁 The export lands outside the project

Gently bring back the file-hygiene habit from Session 1: one folder per episode, clear names.

📏 The measurement is far from target

Check the export settings, or run the file through Auphonic once and measure again.

The phone test tells the truth

It catches what studio headphones hide: harsh "s" sounds, an intro mixed too loud, noise that vanishes on good equipment and roars on cheap earbuds.

🎓 Trainer Tip: Insist on step 7 even when time is short. The phone test is the single habit most likely to survive the course, because it costs nothing and needs no software. Keep one phone with a pair of cheap earbuds at the front of the room as the official "phone test station" for anyone whose own phone is not to hand.

✏️ Exercise — click or tap to expand ▼

▶ ✏️ Try This With Your Group

The Full Run

Every participant finishes a fragment of their own episode. Laptops with the edited episodes open, the chosen intros, headphones, an online loudness meter and phones with earbuds. Display the 7-step checklist on screen for the whole exercise.

Step 1 — finish the edit (15 minutes): work through steps 1 to 4 of the checklist: clean, process, level, and place the intro with a fade and ducking.

Step 2 — export and measure (10 minutes): export the MP3 at the loudness target and verify it with the meter. If the number is far off, fix it and measure again.

Step 3 — the phone test (10 minutes): listen to the exported file on a phone with cheap earbuds. Note anything that sounds different from the studio headphones, and fix the worst problem if there is time.

⏱ Time: about 35 minutes

📋 Alternative — Take-home task: Ask everyone to finish their episode at home using the printed Export Checklist, ticking off each step, and to bring the exported file and its loudness number to the next meeting.

Mini-Showcase and Closing

The course ends where it should: by listening. Each participant plays about 60 seconds of their finished episode on the room speaker, and the group responds with exactly two things.

👍 One thing that works

A specific moment, choice or quality worth keeping.

🔧 One thing to improve

One concrete, doable suggestion — not a list.

About 90 seconds per person, including feedback, keeps the pace brisk and the tone kind. This is the emotional payoff of the whole course, so protect its time: if anything has to be shortened, shorten the theory, not the showcase.

Publishing is a separate topic

Hosting, RSS feeds and distribution platforms are their own subject. This course teaches the craft of making an episode worth publishing.

🎓 Trainer Tip: Split the roles: one trainer moderates the feedback and watches the clock, the other handles playback, so that no time is lost between clips. Close with what participants take home — the handouts, the checklist, the tool list — and with the one-sentence philosophy of the course: good recording first, honest editing second, AI last. A group photo, if the group is up for it, is a good final moment.

✏️ Exercise — click or tap to expand ▼

▶ ✏️ Try This With Your Group

60 Seconds Each

A structured showcase of the finished episodes. A room speaker, a laptop for playback and a timer.

Step 1 — choose (before the showcase): each participant picks the 60 seconds of their episode they are proudest of and notes the start time.

Step 2 — play and respond (about 90 seconds per person): play the fragment; the group offers one thing that works and one thing to improve.

Step 3 — one sentence (3 minutes): close with a round in which everyone says what they will do with their episode next.

⏱ Time: about 20 minutes for 10 participants

📋 Alternative — Take-home task: If there is no time for a live showcase, ask everyone to share a 60-second fragment in the group’s chat or by e-mail, and to reply to two other fragments with one thing that works and one thing to improve.

Key Takeaways

  • AI rescue has honest limits. It handles steady noise and room echo well, but it cannot fix clipping, a voice buried under other sound, or a speaker too far from the microphone.
  • Listen for the artefacts. A watery voice, swallowed consonants and an artificial studio tone mean the AI is working too hard; sometimes the original is better.
  • Music has a job. An intro gives identity; most podcasts need little more. Fade it, duck it under the voice, and remember that silence is a choice too.
  • Length does not create a licence. Use legal music sources and check the licence of every track.
  • Generating is easy; publishing is a licensing decision. For every AI tool, ask two questions: what are the terms, and are we being open about it?
  • WAV to keep, MP3 to publish. Export at about -16 LUFS with the true peak below -1 dB, and verify with a meter.
  • There is a free path. A browser-based editor plus an automatic loudness service covers everything a podcaster needs; the craft transfers.
  • The phone test tells the truth. Finish every episode with the 7-step checklist, and always listen on a phone and cheap earbuds.

Resources & Downloads

Two printable one-page guides. Keep the Export Checklist next to the laptop for every episode, and check the Music & AI Tools card before you publish any music.

  • 📄 The Export Checklist — the 7 finishing steps, the loudness scale, formats and what to do if the number is off target
  • 📄 Music & AI Tools — what music is for, legal sources with their caveats, the two AI generators and the licensing questions to always ask

Tools mentioned on this page