How to apply a broadcast compressor to voice online
- Step 1Open the Speech Leveler and drop your VO — Go to the Speech Leveler and drop the voiceover or narration file (MP3, WAV, FLAC, M4A, OGG, or Opus). FFmpeg 8.1 loads as WebAssembly in-tab; the read stays on your machine.
- Step 2Select the Voiceover preset for the broadcast sound — Open the Preset menu and choose Voiceover — 3:1 over -30 dB. This is the broadcast-leaning curve: it engages compression lower and pushes a hotter output endpoint than the Podcast preset, giving the forward, dense character.
- Step 3Use Podcast for a more natural read — If the spot calls for a warmer, less hyped tone (corporate explainer, audiobook), the Podcast preset (~3:1 over -25 dB, -15 dB endpoint) is gentler and less upfront.
- Step 4Use Interview for a two-voice spot — For a dialogue ad or co-narrated piece, Interview (gentler ~2:1, slower 0.6 s release) keeps both voices even without pumping on the back-and-forth.
- Step 5Run the compressor stage — Process. The tool emits a single compand pass and re-encodes in your source format. This is the compressor stage only — it does not limit peaks or set loudness.
- Step 6Finish the broadcast chain — Send the compressed read to the true-peak-limiter (cap -1 dBTP), then the loudness-normalizer for the delivery spec (e.g. -16 LUFS spoken, -23 LUFS EBU broadcast).
Preset character for voice work
The Voiceover preset is the broadcast-leaning curve. All three are exposed via the single Preset menu — no manual compressor controls.
| Preset | Engages around | Output endpoint | Use it for |
|---|---|---|---|
| Voiceover | -30 dB | -12 dB (hotter, denser) | Broadcast / radio-style VO, ad reads — forward and always-present |
| Podcast | -25 dB | -15 dB | Solo host, audiobook, corporate explainer — natural and warm |
| Interview | -25 dB (gentler ~2:1) | -15 dB, 0.6 s release | Two-voice spots, co-narration, dialogue — even without pumping |
The real broadcast chain (compressor is stage one)
This tool is the compressor only. A finished broadcast deliverable needs the limiter and loudness stages too.
| Stage | Tool | What it sets |
|---|---|---|
| 1. Compress | Speech Leveler (this tool), Voiceover preset | Dynamic range — the forward, even read |
| 2. Limit peaks | true-peak-limiter | True-peak ceiling at -1 dBTP so it never clips on broadcast/stream codecs |
| 3. Set loudness | loudness-normalizer | Integrated LUFS to spec (-16 spoken, -23 EBU R128 broadcast) |
| Optional tone | voice-eq | High-pass + presence boost for clarity before compressing |
Cookbook
Voiceover-focused recipes. The code shows the exact compand curve each preset builds and where the limiter/normaliser take over to finish the broadcast spec.
Radio-style ad read with the Voiceover preset
A 30-second spot that needs to cut through on AM/FM and streaming. Voiceover gives the dense, forward sound.
Preset: Voiceover -af compand=0.05,0.5:0.05,0.5:-90/-90|-70/-65|-30/-20|0/-12:6:0:-90:0.1 Engages at -30 dB, ends hot at -12 dB -> upfront, consistent Then: true-peak-limiter (-1 dBTP) -> loudness-normalizer
Full broadcast chain to EBU R128 -23 LUFS
A deliverable for broadcast TV/radio needs -23 LUFS. The compressor shapes it; the normaliser hits the number.
1. Speech Leveler (Voiceover) -> read-comp.wav 2. true-peak-limiter (-1 dBTP) -> read-limited.wav 3. loudness-normalizer (EBU R128, -23 LUFS) -> deliverable.wav This tool alone does NOT reach -23 LUFS.
Natural corporate narration with Podcast preset
An explainer that should sound human, not hyped. Podcast's gentler curve and -15 dB endpoint keep it warm.
Preset: Podcast -af compand=0.05,0.4:0.05,0.4:-90/-90|-70/-70|-25/-15|0/-15:6:0:-90:0.1 Less forward than Voiceover; preserves a natural read Good for audiobooks and corporate VO
EQ before compress for extra clarity
High-pass the rumble and add presence before compressing so the compressor reacts to a cleaner signal.
1. voice-eq (Male/Female profile: HPF + presence) -> read-eq.wav 2. Speech Leveler (Voiceover) -> read-comp.wav 3. true-peak-limiter -> loudness-normalizer EQ first means the compressor isn't chasing low-end rumble.
Don't expect this to be the limiter
The Voiceover endpoint is hot; on a loud source the peaks can run toward 0 dBFS. compand is not a true-peak limiter.
Symptom: post-compand peaks near 0 dBFS, codec distortion risk Wrong: 'the compressor will catch the peaks' Right: chain the true-peak-limiter after, cap at -1 dBTP
Edge cases and what actually happens
Expecting a built-in limiter or LUFS target
By designThis is the compressor stage only. There is no true-peak limiter and no integrated-loudness target in this tool. The Voiceover preset's hot -12 dB endpoint can leave peaks near 0 dBFS — finish with the true-peak-limiter and loudness-normalizer.
No manual threshold/ratio/attack/release
By designUnlike a DAW compressor, the only control is the three-way preset menu. Each preset is a fixed compand curve. If you need to dial the threshold and ratio by hand, run FFmpeg's compand directly using the curve strings in the cookbook as a base.
Voiceover preset sounds over-hyped
Use PodcastVoiceover is intentionally dense and forward; on a subtle read it can sound pumped. Switch to the Podcast preset (engages at -25 dB, -15 dB endpoint) for a more natural broadcast tone.
Music bed under the VO pumps
Voice-only toolThe curves are tuned for spoken voice. Run the compressor on the dry VO stem, then lay it over the music bed. Compressing the full mix makes the compressor react to the bed and duck it audibly.
Room noise rises with the broadcast density
ExpectedHeavier compression lifts the quiet floor, so booth hiss or HVAC becomes audible. Denoise with ai-noise-reducer before compressing for a clean broadcast read.
Free tier cannot run it
Pro requiredSpeech Leveler is Pro-gated; Free throws 'Speech Leveler requires a Pro subscription.' Processing remains in-browser with no upload on any plan. Pro audio caps are 200 MB / 120 minutes / 10 files.
Long-form narration over Pro limits
RejectedA full audiobook chapter can exceed Pro's 200 MB / 120-minute per-file caps and will be rejected. Split with audio-splitter or use Pro + Media (100 GB, unlimited duration).
Output format isn't broadcast WAV
Convert separatelyThe output keeps your input's format; there is no format selector. If your deliverable spec requires a specific WAV/PCM container and your source was MP3, convert first or after with a sibling such as mp3-to-wav.
Frequently asked questions
Can this give my voiceover a real broadcast sound?
The compressor stage, yes — the Voiceover preset produces the forward, dense, always-present character broadcast voices have. For a complete broadcast deliverable you still chain a true-peak limiter and a loudness normaliser after it.
Which preset is the broadcast one?
Voiceover. It engages compression around -30 dB and ends on a hotter -12 dB output endpoint, which gives the dense, upfront radio sound. Podcast is gentler; Interview is for two voices.
Does it include a limiter?
No. compand levels averages, not inter-sample peaks. Run the true-peak-limiter afterward to cap at -1 dBTP so the read survives lossy/broadcast codecs without clipping.
Does it hit -23 LUFS for EBU broadcast?
Not on its own. It sets dynamics, not loudness. Follow with the loudness-normalizer and choose the EBU R128 preset for -23 LUFS, or -16 LUFS for spoken-word streaming.
Can I set my own compressor parameters?
No — only the three presets are exposed. Each is a fixed compand curve. For hand-tuned threshold/ratio, run FFmpeg's compand directly; the cookbook lists the exact strings these presets build.
Is the client VO uploaded anywhere?
No. FFmpeg 8.1 runs as WebAssembly in your browser, so an unapproved client read never leaves your machine.
Should I EQ before or after compressing?
EQ first. High-pass the rumble and add presence with voice-eq so the compressor reacts to a clean signal, then compress, then limit and normalise.
What formats can I feed it?
MP3, WAV, FLAC, M4A, OGG, and Opus. The output matches the input format; convert with a sibling like mp3-to-wav if your deliverable needs a specific container.
Why does the booth noise come up after compressing?
A denser compressor setting lifts the quiet floor along with the voice. Denoise first with ai-noise-reducer, then compress, for a clean broadcast read.
Will it pump on a dialogue spot?
Use the Interview preset for two voices — its slower 0.6 s release and wider knee ride over the back-and-forth without obvious pumping. Voiceover can pump on rapid dialogue.
How long does a 30-second spot take?
Seconds. compand is a single fast pass with no measurement stage, all on your own CPU, with no upload.
What's the difference between this and podcast-master?
podcast-master runs a full automated chain (denoise, silence strip, normalise, limit) in one click. This tool is just the compressor stage, so you keep manual control over each step of a broadcast chain. Use podcast-master when you want hands-off.
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.