How to improve voice clarity in recordings
- Step 1Confirm you're on Pro — Voice EQ requires a Pro plan (
minTier: pro). On Free it's locked and the audio family caps at 50 MB / 30 minutes. Pro lifts that to 200 MB / 120 minutes per file. - Step 2Add the unclear recording — Drop one file (single-file, no batch). MP3, WAV, FLAC, M4A, OGG, and Opus all decode. A single-voice track gives the cleanest result; EQ each speaker separately if you have isolated tracks.
- Step 3Choose the profile that fits the voice — Start with Male (+3 dB @ 3 kHz) or Female (+3 dB @ 4 kHz, higher high-pass) for general clarity. Use Bright (broadcast) (+4 dB @ 5 kHz) when the voice is still buried, and Warm (gentlest) when the recording is thin or already a little harsh.
- Step 4Process in your browser — FFmpeg applies the high-pass plus two peaking bands as one
-afchain. A typical recording finishes in seconds to under a minute; the first run loads the WASM engine once. - Step 5Listen on the worst-case device — A/B the result on phone speakers or cheap earbuds — that's where clarity problems show. If it's still buried, step up to Bright; if it now sounds harsh or sibilant, the presence boost is too much for this voice — try Warm.
- Step 6If sibilance appears, de-ess after EQ — A presence boost can emphasise harsh S sounds. If the result is sibilant, run the de-esser after this tool to tame the 7 kHz region without dulling the clarity you just added.
Which profile for which clarity problem
Pick by the symptom you hear. All four profiles use the same three-stage architecture; only the frequencies and gains differ. Values read directly from the processor.
| If the voice sounds… | Use profile | Why (exact curve) |
|---|---|---|
| Boxy / far away, normal pitch | Male | highpass=f=80, -2 dB @ 300 Hz, +3 dB @ 3 kHz |
| Buried, higher-pitched voice | Female | highpass=f=100, -2 dB @ 300 Hz, +3 dB @ 4 kHz |
| Still dull after Male/Female | Bright (broadcast) | highpass=f=80, -1.5 dB @ 250 Hz, +4 dB @ 5 kHz |
| Thin or already a bit harsh | Warm | highpass=f=70, -1 dB @ 350 Hz, +2 dB @ 2.5 kHz |
The clarity bands explained
Why these specific frequencies improve intelligibility for spoken voice.
| Band | Action | Effect on clarity |
|---|---|---|
| Below 70-100 Hz | High-pass (remove) | Strips rumble that masks the voice and wastes headroom |
| 250-350 Hz | Peaking cut (-1 to -2 dB) | Removes the boxy congestion that makes a voice sound 'covered' |
| 2.5-5 kHz | Peaking boost (+2 to +4 dB) | Lifts consonant detail — the band the ear uses to parse words |
Tier limits
Voice EQ is a Pro feature. Per-file limits; single file per job.
| Plan | Max size | Max duration | Access |
|---|---|---|---|
| Free | 50 MB | 30 min | Locked |
| Pro | 200 MB | 120 min | Yes |
| Pro-media | 100 GB | Unlimited | Yes |
| Developer | 100 GB | Unlimited | Yes |
Cookbook
Real clarity rescues, each with the exact FFmpeg chain so you can see what changed.
Zoom interview that sounds boxy and distant
Male profile. The 300 Hz cut pulls the voice out of the box and the 3 kHz boost brings it forward.
Profile: Male highpass=f=80, equalizer=f=300:t=q:w=1:g=-2, equalizer=f=3000:t=q:w=1:g=3 interview.wav -> interview-eq.wav
Higher-pitched lecturer who's hard to follow
Female profile lifts presence at 4 kHz and uses a 100 Hz high-pass suited to the higher fundamental.
Profile: Female highpass=f=100, equalizer=f=300:t=q:w=1:g=-2, equalizer=f=4000:t=q:w=1:g=3 lecture.mp3 -> lecture-eq.mp3 (re-encoded libmp3lame)
Still muddy after a first pass
Bright (broadcast) is the most forward profile — biggest presence lift (+4 dB) at 5 kHz. Use it when standard profiles aren't enough.
Profile: Bright (broadcast) highpass=f=80, equalizer=f=250:t=q:w=1:g=-1.5, equalizer=f=5000:t=q:w=1:g=4
Clear but now sibilant — chain a de-esser
Adding presence can make S sounds harsh. EQ for clarity, then de-ess to tame the result.
Step 1: voice-eq (Bright) -> clearer but harsh S
Step 2: /audio-tools/de-esser (intensity ~0.6)
-> tames 7 kHz sibilance, keeps the clarityDon't try to fix loudness here
EQ improves tone, not level consistency. If quiet words still get lost, level and normalise after EQ.
1. voice-eq (clarity / tone) 2. /audio-tools/speech-leveler (even out level) 3. /audio-tools/loudness-normalizer (-16 LUFS)
Edge cases and what actually happens
Louder, but still muddy
Wrong toolIf you turned the file up and it's still unclear, volume was never the problem — frequency balance is. This EQ targets the muddy 300 Hz and the presence 3-5 kHz bands; that's what actually changes clarity. Don't reach for a volume tool to fix tone.
Clarity boost amplified the hiss
Order issueThe presence boost lifts everything in its band, including noise. If the result is hissier, you EQ'd before denoising. Run the ai-noise-reducer first, then EQ.
Voice is now harsh or sibilant
Expected trade-offMore presence can over-emphasise S and T sounds. Switch to the gentler Warm profile, or keep the brighter profile and run the de-esser afterwards.
No custom EQ bands available
By designThere are no manual sliders — just the four-profile picker. Each is a fixed curve. For a fully parametric EQ you'd need a DAW; this is a curated speech chain.
Free user can't run it
Pro requiredThe tool is gated at Pro despite 'free' in the URL. Free caps the audio family at 50 MB / 30 minutes and locks this tool.
Heavy room reverb won't clear up
Room limitEQ shapes tone but cannot remove reflections. A very reverberant recording will sound clearer but still washy. Re-record with treatment or closer mic placement for a real fix.
Two very different voices in one file
SuboptimalA single profile can't be ideal for both a deep and a high voice at once. EQ isolated tracks separately before mixdown for the best clarity on each.
Recording is over the duration limit
400 — duration exceededPro allows 120 minutes per file. A longer recording is rejected even under 200 MB. Split with the audio-splitter, EQ each part, then rejoin.
MP3 source, slight extra loss
ExpectedEQ re-encodes. An MP3 in becomes a re-encoded MP3 out (libmp3lame). Negligible for speech, but EQ a WAV/FLAC master if you have one.
Whispered or extremely quiet speech
Noise riskA very quiet voice has a high noise-to-signal ratio; boosting presence raises the noise too. Denoise first and consider leveling before EQ for the cleanest clarity.
Frequently asked questions
How do I make a voice recording clearer?
Cut the boxy low-mid around 300 Hz and boost the presence range around 3-5 kHz, after high-passing out rumble. This tool does all three in one pass per the profile you pick.
Why does turning the volume up not help?
Volume scales everything equally, including the mud. Clarity is about the balance between the boxy 300 Hz region and the presence region — that's what EQ changes.
Which profile is clearest?
Bright (broadcast) has the biggest presence boost (+4 dB at 5 kHz), so it's the most forward. Male and Female are good general defaults; Warm is gentlest.
Can I pick exact frequencies?
No — the only control is the four-profile picker, each a fixed curve. It's a curated speech EQ, not a parametric one.
Should I denoise first?
Yes. Denoise before EQ so the presence boost doesn't amplify hiss. Use the ai-noise-reducer, then this tool.
What if it gets sibilant?
Run the de-esser after EQ, or use the gentler Warm profile. A presence boost can over-emphasise S sounds.
Is the recording uploaded?
No. It's processed in your browser via FFmpeg WASM and never leaves your device.
Is it free?
It's free of cloud and upload, but the tool requires a Pro plan. Free doesn't include it.
What formats are supported?
MP3, WAV, FLAC, M4A, OGG, and Opus decode. Output uses the same container as the input.
Will it fix a bad room?
Only partially. The high-pass and low-mid cut help, but heavy reverb can't be EQ'd away. Treat or re-record the source.
Can I batch several files?
No — one file per job. For a multi-file mastering chain, use podcast-master.
How big and long a file can I use?
Up to 200 MB and 120 minutes on Pro; unlimited duration on Pro-media and Developer.
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.