From Idea to Episode

The Overview page is the episode's brief. Set a working title, the episode number and a target length, and write a 1–2 sentence synopsis (the premise — what this episode is about). Stuck? Press Generate synopsis and edit what the AI drafts. Set a record date here too, then move the episode to scheduled.

On Planning → Speakers, add the person you're interviewing, mark them host or guest, and track where they're up to (invited → confirmed → scheduled). Record their talent release here, which is the formal version of asking permission to record and publish (the consent step below). You can send them a briefing from here too.

Good interviews come from good research. Two tools:

• Research your guest: open their speaker profile and press Research Speaker. Tattle web-searches them and drafts a background bio. Read it, keep what's true, fix what isn't.
• Research the topic: on the Research tab, ask the AI for background on the subject, your guest's history or key moments (e.g. "what are the big turning points in Grandma's life?"). Save the useful answers as research blocks and drag them into order. These become your prep notes.

Always read what the AI gives you with a critical eye: it's a fast first draft, not the truth. Check anything important against what your guest actually tells you.

On Planning → Questions, build the running order of questions you'll actually ask. Type your own, or press AI Suggest for a first batch: it reads your guest's research and avoids repeating questions from their other episodes. Refine any single question with the sparkle button (e.g. "make it more personal"), add a topic tag, and drag them into the order you want. During the interview you can tick each one Asked and rate the answer, so the runsheet doubles as your checklist.

• Power: the H5 has no internal battery — put in 2× AA batteries (or run it from USB). Fresh batteries last about 15 hours.
• SD card: with the recorder off, slide the SD card in (face up) on the right-hand side. Turn on, then format the card in the menu so it's clean and compatible.
• Recording mode: choose Multi File and set the format to WAV, 48 kHz / 24-bit (WAV is an uncompressed, top-quality audio file; 48 kHz / 24-bit is its quality setting, and it matches what you'll use in Audition).
• Plug in the mics: connect each ATR2100x with an XLR cable (the standard 3-pin microphone cable) into Input 1 and Input 2 at the bottom, one mic per person, each on its own track. (Ignore the X/Y mic on top, the recorder's built-in stereo mic, for interviews.)

• The ATR2100x is a dynamic mic, so it does not need phantom power. Leave 48V off for both mics: the H5's preamps have plenty of gain for these mics on their own, so you do not need the Cloudlifter here. (The Cloudlifter and 48V phantom come in later, only on the desk setup with the Scarlett on Day 4.)
• Arm the tracks: press the Track buttons (1 and 2) so they glow red. Only armed tracks record.

• Each input has its own gain dial (under the little protective bar). No buttons needed — just turn the dial for that channel.
• Have each person talk at normal volume. Turn their dial up until the loudest bits peak around −12 dB (decibels, the unit on the level meter) — strong and clear, never slamming into the top (0 dB is distortion you can't fix).
• Build in a real safety net. The H5's Backup Record only copies the top X/Y mic (the L/R track), not Inputs 1 & 2, so it will not protect your interview mics. Instead, run a phone voice memo alongside as a genuine second recording. Never trust a single recording of an interview you can't redo.
• Plug the ATR-M70x headphones into the H5's headphone socket and actually listen while you set levels. Your ears are the best meter.

• Open a Room: in Descript on the Mac, start a Room and copy the join link. Your guest opens that link in a web browser, so they need no app, no account and no special gear.
• Your end is the desk chain: Mic → Cloudlifter → Scarlett → Mac, with headphones on. This is the one time you use the Cloudlifter and Scarlett to capture an interview.
• Point Descript at the Scarlett: in Descript's audio settings, set the input to the Scarlett (not the Mac's built-in mic) and the output to your headphones, so it records your good mic and you hear the call cleanly.
• Give yourself the best line: use a wired internet connection if you can; if not, sit right next to the router. Close other apps and tabs so the Mac isn't struggling.

The real skill of remote recording is getting someone else's setup right from a distance, kindly and clearly. Talk them through it before you record:

• Headphones on — the golden rule. Headphones stop your voice leaking out of their speakers and back into their mic, which is what causes echo. Any headphones or earbuds will do; laptop speakers are the one thing to avoid.
• The quietest room they have, door shut. Turn off fans, the TV and notifications, put phones on silent, and settle pets. A noisy room is far easier to prevent than to fix afterwards.
• Get close to the mic. Whatever they have — a headset, earbuds with a mic, a USB mic, or just the laptop — closer and slightly off to the side beats far away and head-on. A headset mic beats the laptop's own.
• A solid browser: Chrome or Edge on a laptop is the safe choice, plugged in to power so it doesn't sleep mid-interview.
• Check their level: ask them to talk at normal volume and watch their meter in Descript — strong and clear, never slamming the top. Ask them to roughly match your loudness, so one of you isn't tiny and the other huge.
• Confirm local recording is on for each person (Descript records each track on its own device — that is what protects you from the internet).

• Test first: record a 20–30 second test with both of you talking, then play it back. Listen for echo (almost always a headphones problem), background noise, clipping, and whether your two levels sit together.
• Consent on the record: once you're rolling, before the real questions, ask your guest out loud whether they're happy to be recorded and for the episode to be shared publicly, and capture that "yes" at the very start — exactly as for the in-person interview.
• Run the interview from your runsheet: a calm intro, one question at a time, and note the times of the best answers. Let them finish before you speak — it's kinder, and it makes the editing far cleaner.
• Watch it the whole way: keep half an eye on the connection and on both record indicators, so you catch a problem in minute two, not at the end.
• Backup: the safety net is that each person's audio records locally, in full quality, even if the connection stutters — that is the whole point of Descript Rooms. Before anyone leaves, confirm both people's tracks have finished uploading (Descript uploads them in the background); don't end the Room until they show as done.
• Hand-off: the finished recording lands straight in Descript, ready for your rough cut on Day 3. Upload it to the episode in Tattle too.

• Speaker diarisation labels who said what, so you can see the conversation as a script.
• AI Correction suggests fixes for misheard words — you accept or reject each one, so you stay in control and learn to spot errors.
• Version history means you can always compare changes and undo. Nothing is ever lost.
• This corrected transcript powers everything later — the script, shownotes, grab quotes and metadata all build on it.

Descript reads the transcript and flags the filler words that pepper natural speech — "um", "uh", and, if you let it, "you know", "like", "so", "I mean". You can then strike them out in bulk or step through them one at a time, and the audio is cut to match. Used well it tightens a ramble; used bluntly it makes a person sound like a robot.

• Do the content edit first. Cut whole tangents and dead sections before you hunt fillers — there is no point polishing "um"s out of a passage you are about to delete.
• Cut the easy ones in bulk. "Um" and "uh" almost never carry meaning, so removing all of them is usually safe. Turn just those on and let Descript clear them in one pass.
• Review the rest by hand. "You know", "like", "so" and "I mean" are discourse markers — they shape natural speech, and stripping every one leaves the talk clipped and stilted. Step through these in context and keep the ones that sound human.
• Listen across every cut. Removing a filler makes a splice, two pieces of audio joined where the word used to be. Check the join hasn't clipped the start of the next word, swallowed a needed breath, or left an audible bump; nudge the cut or leave a tiny gap if it has.
• Leave some in. A perfectly filler-free track sounds over-produced. A thoughtful "um…" before a hard answer can be the most human moment in the episode. Aim for clarity, not sterility.

Pitfalls: trusting "Remove all" blindly — the detector mis-tags a meaningful "like" or misses a stretched "uhhh", so always review; cutting so tight the speech feels rushed and breathless; and leaving clicky edits at the joins. A good test: play a minute to someone who wasn't there. If it still sounds like a normal person talking, you have it right.

• Noise Reduction — capture a "noise print" from a second of silence (room tone), then tell Audition to subtract that hum/hiss everywhere. Go gently to avoid a "underwater" sound. iZotope RX is the specialist version of this.
• Healing / spectral repair — paint out one-off clicks, mouth noises or a stray bump.
• Parametric EQ (equalisation, precise control of the tone) — roll off low rumble, add a little "presence" so speech is clear, tame any harshness.
• Compression — gently evens out the loud and quiet bits so the voice stays steady.
• Mastering & loudness — the final step. Podcasts are delivered to a target loudness, not just a peak level, so the whole episode sits at a steady volume. Watch the TC Electronic loudness meter and bring the integrated loudness of the finished episode to −16 LUFS (the house standard for this show), keeping the true peak under about −1 dBTP so nothing distorts. Master in iZotope Ozone 12 and use its limiter, the Maximizer (which raises the overall level and stops the loudest peaks from clipping), to bring it up to −16; then re-check the integrated reading across the whole episode. The two interview tracks are summed to mono for delivery. You'll see −16 LUFS quoted as the streaming/stereo figure and −19 for mono in a lot of guides; both are fine — what matters is picking one target and hitting it consistently, and ours is −16.

1. Repair first. Clean before you shape — there is no sense EQ-ing or compressing a hiss you could simply remove. Take out steady hum and hiss, and one-off clicks and mouth noises, with iZotope RX or Audition's Noise Reduction. If a recording sounds boxy or distant, reach for de-reverb, never added reverb. Pitfall: over-cleaning leaves a watery, underwater warble, so work in small amounts and take your noise print from real room tone.
2. High-pass filter. Roll off the subsonic rumble that lives under a voice — footsteps, traffic, air-conditioning, desk thumps. Sweep the high-pass filter up from the bottom until the voice just begins to thin, then ease back; around 80 Hz suits most voices, lower for a deep one. Pitfall: cutting too high hollows the voice out and makes it sound small.
3. Subtractive EQ. Shape the tone by cutting problems before boosting anything. Mud sits around 200–400 Hz, boxiness around 300–500 Hz, harshness around 2–5 kHz. To find a trouble spot, boost a narrow band and sweep until it sounds its worst, then cut there instead. The rule: cut narrow, boost wide. Pitfall: stacking boosts makes a voice brittle and louder rather than better.
4. De-ess. Tame sibilance — the harsh "sss" and "sh" energy around 5–9 kHz — with a de-esser, and only as much as the words need. Pitfall: too much gives a lisp; and because compression can push sibilance back up, you sometimes de-ess again afterwards.
5. Ride the level. Before the compressor does anything, even out the big volume swings. A level-rider such as Vocal Rider automatically rides the volume fader the way a human engineer would, holding the voice on a steady target so the compressor barely has to work; with no plug-in, Audition's volume envelopes do the same job by hand. Pitfall: in silent gaps a rider can lift room noise, so use it after de-noising, or pair it with a gentle gate.
6. Compression. Now even out what remains, so every word sits in the same loudness window. Start gentle: a ratio around 2:1 to 3:1, a medium attack, an auto or medium release, aiming for about 3–6 dB of gain reduction on the loudest words, then add make-up gain to restore the level. Two light compressors in series sound smoother than one heavy one. Pitfall: too much pumps, sounds lifeless, and makes breaths and background jump out. A stubborn voice can take multiband compression, which squeezes the low, middle and high bands separately.
7. Gate or expander. Drop the level between words to tuck away breaths and room tone with a gentle downward expander (a soft "noise gate"). Pitfall: a hard gate chops the tails off words and chatters on quiet speech, so use soft expansion and never silence the breathing completely — a little breath sounds human, none sounds robotic.
8. Tone & presence EQ. After compression, a small, wide lift around 3–6 kHz adds clarity, and a touch above 10 kHz adds "air". Pitfall: chasing brightness stacks harshness; if it starts to hiss, you have gone too far.
9. Saturation (optional). A hint of saturation adds harmonic warmth and body, and helps a thin voice feel fuller. Pitfall: more than a hint is just distortion.
10. Limiter. A limiter catches stray peaks and lifts the final level a touch. Let the level-riding and compression do most of the loudness work; the limiter only shaves the last peaks. Pitfall: slamming a limiter to get loud crushes the life out of speech and adds distortion.
11. Master to −16 LUFS. Set the integrated loudness to −16 LUFS with the true peak under about −1 dBTP, using Ozone's Maximizer and the TC Electronic meter. Judge loudness by ear at a fixed monitoring level — loud always seems better at first. Pitfall: do not chase loudness past distortion; loudness normalisation turns every show to a similar level on the platforms anyway.

You don't upload directly to Spotify or Apple. Instead:

• The episode goes to a hosting platform — for this show that's Simplecast. Tattle pushes the audio, title, description and shownotes straight there for you.
• Simplecast creates an RSS feed (RSS stands for Really Simple Syndication) — a constantly-updating list of all your episodes.
• Apple, Spotify and other apps subscribe to that feed and automatically pull in each new episode.
• That's why consistent, accurate metadata (title, episode number, description) matters — it's what listeners see everywhere.