How to level podcast voice loudness — browser compressor
- Step 1Open the Speech Leveler and drop your episode take — Go to the Speech Leveler tool and drop your voice file. It processes audio files — MP3, WAV, FLAC, M4A, OGG, and Opus. The FFmpeg 8.1 WebAssembly engine loads in-tab and decodes the file locally; nothing is sent to a server.
- Step 2Leave the preset on Podcast for a solo voice take — The only control is a Preset dropdown with three options. For a single-host episode keep the default Podcast — 3:1 over -25 dB. It is tuned for one consistent mic and applies a soft-knee curve that levels without sounding squashed.
- Step 3Switch to Voiceover only if your delivery is very dynamic — If your read swings hard between whisper and shout, the Voiceover — 3:1 over -30 dB preset starts compressing lower and adds a touch more make-up via its transfer curve. Use Interview only when you have two or more voices at different distances.
- Step 4Run the leveler — Click process. The tool builds a single
-af compand=...pass and re-encodes with the codec that matches your file's extension. A 40-minute take is typically done in well under a minute on a modern laptop CPU. - Step 5Download and A/B against the original — Listen on a phone speaker, not just headphones — that is where uneven levels expose themselves. The leveled file should hold a steadier loudness with the quiet lines now audible.
- Step 6Send the leveled file to the loudness normaliser — Compression set the dynamics; it did not set the loudness target. Drop the leveled file into the loudness-normalizer for a 2-pass EBU R128 pass to -16 LUFS (podcast) before you publish.
The three Speech Leveler presets
These are the only options the tool exposes — a single Preset dropdown. Each preset maps to a fixed FFmpeg compand transfer curve; there are no separate threshold, ratio, attack, or release sliders.
| Preset | Best for | Attack / release | Character |
|---|---|---|---|
| Podcast (default) | Solo episode, one consistent mic | 0.05 s / 0.4 s | Soft-knee, roughly 3:1 above -25 dB — tightens loud peaks, lifts quiet lines, transparent on a single voice |
| Voiceover | Very dynamic narration / dramatic reads | 0.05 s / 0.5 s | Starts levelling lower (around -30 dB) with more make-up gain — denser, more upfront, for swing-heavy delivery |
| Interview | Two or more voices at different distances | 0.1 s / 0.6 s | Gentler, closer to 2:1 with a slower release — pulls a far guest and a close host into the same range without pumping |
What this tool does vs. what to use instead
Speech Leveler only smooths dynamic range. For loudness targets, peaks, EQ, or file size, the right tool is a sibling.
| Your goal | Right tool | Why not Speech Leveler |
|---|---|---|
| Hit -16 / -14 LUFS for Spotify / Apple | loudness-normalizer | Compand has no LUFS target; it tightens range, it does not set integrated loudness |
| Stop the file clipping on upload | true-peak-limiter | Compand levels averages, not inter-sample true peaks; a limiter caps at -1 dBTP |
| Clean up tonal balance / muddiness | voice-eq | Compand is amplitude-only; EQ shapes frequency (high-pass, presence boost) |
| Shrink the MP3 file size for delivery | audio-compressor | That tool targets file size in MB; this one is dynamic-range compression, a different meaning of 'compress' |
Cookbook
Concrete before/after for a solo-podcast voice take. The code blocks show the FFmpeg compand argument the preset builds and the audible result — there are no numeric sliders to set, only the preset choice.
Default Podcast preset on a wandering solo take
A single-host episode where the intro is loud and the mid-episode asides drop 10-12 dB. The Podcast preset is the right default — it lifts the quiet band and tames the loud band with a soft knee.
Preset: Podcast (default) FFmpeg builds: -af compand=0.05,0.4:0.05,0.4:-90/-90|-70/-70|-25/-15|0/-15:6:0:-90:0.1 Before: intro at ~-12 dB, asides at ~-26 dB (14 dB spread) After: intro and asides land within ~5-6 dB of each other Format: WAV in -> WAV out (matches input extension)
Compress here, then normalise — the publish chain
The leveler does dynamics; it does not set loudness. For a Spotify/Apple-ready episode, run this first, then the EBU R128 normaliser.
Step 1 (this tool): Podcast preset -> episode-leveled.wav Step 2 (loudness-normalizer): preset Apple Podcasts (-16 LUFS) 2-pass loudnorm measures, then corrects + true-peak limits Result: even dynamics AND a correct -16 LUFS integrated target Do NOT expect this tool alone to reach -16 LUFS.
Voiceover preset for a dramatic solo read
A storytelling episode with whispered build-ups and shouted reveals. The Podcast preset leaves it still a bit dynamic; Voiceover clamps harder.
Preset: Voiceover FFmpeg builds: -af compand=0.05,0.5:0.05,0.5:-90/-90|-70/-65|-30/-20|0/-12:6:0:-90:0.1 Whispers lifted further, reveals held down -> denser, upfront Use Podcast first; switch to Voiceover only if it sounds too dynamic.
FLAC archive take stays lossless through the pass
You keep masters in FLAC. The output extension follows the input, so a FLAC drop stays FLAC (re-encoded with the flac encoder).
Input: episode-master.flac Preset: Podcast Output: episode-master-leveled.flac (FLAC, lossless container) There is no format dropdown for this tool; the output extension always mirrors the input file's extension.
Don't double-compress an already-mastered file
If you already ran podcast-master (which includes leveling and normalisation), running the leveler again over-squashes the voice.
Wrong: master -> podcast-master output -> Speech Leveler again Symptom: lifeless, pumping, breath noise pumped up loud Right: raw take -> Speech Leveler -> loudness-normalizer OR raw take -> podcast-master (one-click chain) only
Edge cases and what actually happens
Free tier tries to run the leveler
Pro requiredSpeech Leveler is a Pro-tier feature. On the Free tier the processor throws 'Speech Leveler requires a Pro subscription.' before any audio is touched. The tool still runs 100% in your browser with no upload — the gate is on the subscription, not on where processing happens. Free covers audio files up to 50 MB / 30 minutes / 1 file across the family; Pro raises that to 200 MB / 120 minutes / 10 files.
File over the Pro 200 MB / 120-minute limit
RejectedPro audio limits are 200 MB per file and 120 minutes per file. A long-form episode that exceeds either is rejected. Pro + Media lifts both to 100 GB / unlimited duration. The duration cap is separate from the byte cap — a short but very high-bitrate WAV can hit 200 MB before 120 minutes, and a long low-bitrate MP3 can hit 120 minutes before 200 MB.
Expecting it to reach a LUFS target
By designCompand smooths dynamic range; it has no integrated-loudness target. The leveled file will sound more consistent but is not guaranteed to be at -16 LUFS. Always follow with the loudness-normalizer for a publish-ready loudness number.
Background hiss gets louder after leveling
ExpectedCompression lifts quiet passages — including room hiss, fan noise, and air conditioning in the gaps. This is correct compander behaviour, not a bug. Denoise first with the ai-noise-reducer (RNNoise), then level, so you are not amplifying noise into the quiet band.
Output clips or peaks above 0 dBFS
Use a limiterThe presets include make-up gain in their transfer curves; an already-hot input can push peaks toward 0 dBFS after leveling. The leveler is not a peak limiter. Run the true-peak-limiter afterward to cap at -1 dBTP and stay clip-safe through lossy encoding.
Music bed under the voice gets pumped
Voice-only toolThe compand curves are tuned for spoken voice. Applied to a full mix with a music bed, the compressor reacts to the loudest element and can audibly duck the bed. Level the voice stem before you add the bed, or master the final mix with a gentler tool.
Video file dropped in
Audio in, audio outSpeech Leveler is registered for audio input and outputs audio. To level the voice in a video, first extract the audio with a sibling such as video-to-wav, level the WAV, then re-mux. The dropzone copy mentions video for the family, but this tool's pipeline produces an audio file.
Already-compressed delivery MP3 re-leveled
CautionRunning the leveler on a finished, already-mastered MP3 re-encodes lossy-to-lossy and stacks compression on top of compression — thin, pumping results. Level the rawest take you have (ideally WAV/FLAC), not the published export.
Frequently asked questions
Is my episode uploaded anywhere?
No. The Speech Leveler runs FFmpeg 8.1 compiled to WebAssembly inside your browser tab. The audio is decoded, leveled, and re-encoded locally — it never reaches a server, so an unreleased episode stays private on your machine.
Does this set my podcast to -16 LUFS?
No. This tool compresses dynamic range; it does not target integrated loudness. After leveling, run the loudness-normalizer and choose the Apple Podcasts preset (-16 LUFS) or Spotify/YouTube (-14 LUFS) for a proper 2-pass EBU R128 target.
Which preset should a solo podcaster use?
Start with the default Podcast preset (about 3:1 over -25 dB, 0.05 s attack / 0.4 s release). It is tuned for a single consistent mic. Only switch to Voiceover if your delivery is unusually dynamic, or Interview if you have more than one voice.
Can I set my own threshold, ratio, attack, and release?
No. The only control is the Preset dropdown with three fixed curves (Podcast, Voiceover, Interview). Each preset maps to a complete FFmpeg compand transfer curve internally; there are no separate numeric sliders for threshold, ratio, attack, or release.
What output format do I get?
The output extension matches your input. Drop a WAV and get WAV; drop an MP3 and get MP3; FLAC stays FLAC. There is no format selector on this tool. If you need a different container, convert afterward with a sibling such as wav-to-mp3.
What file types can I drop in?
Audio files: MP3, WAV, FLAC, M4A, OGG, and Opus. For a video, extract the audio first with video-to-wav, level it, then re-mux into your video.
How big a file can I level?
On Pro, up to 200 MB and 120 minutes per file. Free covers up to 50 MB / 30 minutes but does not unlock this tool (it is Pro-gated). Pro + Media lifts the cap to 100 GB and unlimited duration for full-length productions.
Why did the background hiss get louder?
Compression lifts quiet sections, and your room noise lives in those sections. That is expected. Denoise first with the ai-noise-reducer, then level, so the quiet band is clean before it gets boosted.
Is this the same as the audio-compressor tool?
No — they share the word 'compress' but do different jobs. Speech Leveler is dynamic-range compression (loud/quiet evening). audio-compressor is file-size compression — it targets a smaller MB output. Pick by what you mean by 'compress'.
Should I level before or after editing?
After your cut is locked but before normalisation. Edit and remove dead air first (try silence-stripper), then level the full take, then normalise. Leveling the final assembled voice gives the most consistent result.
Will it pump or breathe on my pauses?
The Podcast preset's 0.4 s release is slow enough to ride over natural pauses without obvious pumping. If you hear pumping it usually means the input was already compressed; level the rawest take instead of a finished export.
Can I do a whole season in one go?
Speech Leveler processes one file per run. On Pro you can keep up to 10 files queued in the family workflow, but each leveling pass acts on a single recording. For a hands-off full chain across episodes, the podcast-master one-click tool may fit better.
Privacy first
Every JAD Audio tool runs entirely in your browser via FFmpeg (WebAssembly) and RNNoise. Your audio files never leave your device — verified by zero outbound network requests during processing.